Horizon Academic Research Journal Vol. 1 First Edition - Book View

The Horizon Academic
Research Journal
A Collection of
Undergraduate-Level Research,
Composed by
High School Students
Volume 1 2020
TABLE OF CONTENTS
The Effects of the Coronavirus Pandemic on Student Learning Anika Khurana 01
Melanoma Diagnosis using Convolutional Neural Networks Ansh Chaurasia 14
Copyright 2021, Sunrise International Education Inc.

Behavioural Economics and Education: Exploring possible Arzoo Usgaonkar 29
methods by which heuristics could be used to improve the
Publisher performance of students in the International Baccalaureate
Primedia E-launch LLC Diploma Program (IBDP)
PO Box 2727, Orlando, Florida, 32802, United States
Oncogenic Viruses: A Potential CRISPR Treatment Neha Matai 50

ISBN 978-1-63821-125-9
The Horizon Academic Research Program Predicting NBA Playoffs Using Machine Learning Sean Liu 74
A Project of Sunrise International Education Inc.

Email: Contact@HorizonInspires.com
Extending the Drift Diffusion Model to the Cognitive Realm Xiaoyu Wu 88
Website: www.horizoninspires.com
Anxiety and Procrastination: What is the Association? Hriday Nayak 109

The Horizon Academic Research Journal is published by
Sunrise International Education Inc. The 2020 edition is Volume 1.
Prioritizing missense mutations in BDNF to predict variant Srihas Rao 135
pathogenicity in Alzheimer’s Disease (AD)
Copyright ©2021, by Sunrise International Education Inc., A Perspective on Novel Proteins Maggie Lau 152
102 Christopher Columbus Dr., Ste. 1101, Jersey City, NJ 07302, USA. All rights reserved.
No part of this publication may be reproduced, stored in retrieval system,
or transmitted in any form or by any means, A Deconstructive Approach to Hippocampal Neurogenesis as a Harshitha Valluri 165
electronic, mechanical, photocopying, recording, Function of Aerobic Exercise

or otherwise, without the prior written permission of the publisher.
The Neural Relationship Between Smooth Movement and Curie Cha 179
Musical-Motor Entrainment and Applications in Musical
Rehabilitation
A NATO intervention: The Right Choice for Libya Anya Nedungadi 187
What, if Anything, is Wrong with Surveillance Capitalism? Aaron Chen 196

Foreword
The connectedness of today’s world enables us to unlock the boundless

potential of human ingenuity. It is in this spirit that The Horizon Academic Research
Program connects talented high school students from all over the world with
professional researchers to work on meaningful academic research. The Horizon
Academic Research Journal showcases the selected high quality work produced by
high school students in the The Horizon Academic Research Program.
This journal issue represents diversity and excellence in research from our
Summer 2020 term. Students from 27 countries, from our own zip code in New
Jersey, to as far away as Vietnam, India, Armenia, and China, worked with researchers
from 12 universities to produce scholarship on topics ranging from Bioinformatics
to International Relations. By conducting the program online, we were able to cross
physical boundaries that would have otherwise made such work impossible.
Out of more than 1,200 of applicants, 173 students were accepted and enrolled into
the program. These students worked either individually or in small groups with
researchers, with the goal of producing an impactful research project. Some students
produced original research, while others filled gaps in the current literature using
knowledge gained over the course of their time at Horizon Academic. Of these
students, 12 were selected to publish their work in this issue.
Despite being written by high school students, these papers display a level of
thoughtfulness, analytical rigor, and ambition more typical of undergraduate students.
Some papers are approachable introductions to fields that are difficult for the average
person to access, while others cause us to question our understanding of ideas that we
may think we are familiar with.
We are pleased to make this Volume available to the public and hope that this
provides a window, not only into the skill and ability of each author, but also into the
abilities of high school students to do meaningful research without the use of physical
lab spaces. It is our pleasure and honor to share this work with you.
Ani Nadiga
Horizon Academic Research Program, Project Advisor
Editor, Horizon Academic Research Journal, Vol. 1
first identified as COVID-19 in December 2019 in Wuhan, China. On January
20th, 2020, the United States recognized its first case in Washington [CDC20]
The virus was officially declared a pandemic on March 11th, 2020 [Org20]
The Effects of the Coronavirus Pandemic on The exponential spread of this contagion has alerted many officials to take
drastic measures for the safety and well-being of their nations’ populations. To
Student Learning disrupt human-to-human contact, many counties, states, and countries have
announced lockdowns, ordering civilians to isolate and shelter in place.
[Org20] The global pandemic has led to a severe health and socioeconomic
Anika Khurana ∗ crisis. Immunocompro- mised and older populations in specific are
compelled to take extreme precautions as they are at a greater risk of death by
September 30, 2020 COVID-19 [fDCP20]. An estimated 195 million jobs will be lost, and 800
million people will not be able to meet basic needs [Pro20]. This
unprecedented disaster has left millions in unstable living conditions and
food insecure. Concerning the educational impact, China and Mongolia were
Abstract
the first countries to close schools on February 17th, 2020, affecting one
The Coronavirus pandemic (COVID-19) has spread globally with catas- million learners. By April 4th, 2020, 192 countries implemented school
trophic effects on the world, severely impacting lives, jobs, and the ed- closures, affecting 1.6 billion learners, accounting for 91.2Certain popu-
ucation of students. The health crisis impact has forced a shift towards
lations of students have been particularly affected by the transition to remote
online learning, transforming the delivery of education, and disrupting the
schooling of every student. This paper looks at the comparison between
learning. Young children, for example, are faced with the disadvantage of being
traditional and online learning and the effects they have on social inter- technologically illiterate, requiring additional assistance. Learners with learning
actions, motivation, and stress on students. Although every learner has disabilities are also an example of an affected population deprived of the
been negatively affected by this sudden adjustment to remote learning, proper help needed to continue a successful education. This paper focuses on
learners from underprivileged societies have been more vulnerable, facing the additional challenges of online learning that affect underprivileged
disproportionate losses in their education. Online learning has become the students from lower socioeconomic societies including the deep digital divide
new “normal” as the virus’s effect has shown to be longer than predicted. and lack of many crucial elements to make learning more efficient. The
As discussed in this paper, it is crucial to take collaborative actions and infectious disease has impacted everyone in different ways. Individuals are
policy changes to create online learning a more engaging and effective having to use their cognitive resources on factors that could include one’s
delivery method of education.
health, employment, education, or lifestyle, taking away from their
concentration, affecting their performance and well-being. The sudden impact
1 Introduction of the Coronavirus pandemic has left everyone unprepared and isolated,
leading to an inadequate education delivery system. The movement towards
The Coronavirus pandemic, or COVID-19, has revolutionized the educational remote learning has been stressful to many students as important academic
system and its delivery to students across the world. The airborne disease exams and events including the SATs/ACTs and AP exams have been
transmits deadly droplets through close contact between people. Consequently, canceled, postponed, or relocated online. This paper explores the causes of
social distancing is enforced by governments across the world, leading to school increased anxiety and loss of motivation that learners are experiencing which is
shutdowns and an immediate switch to online education. Although online learn- negatively reflected in their performance. To understand the radical changes
ing is not a novel concept, the virus acted as a catalyst, forcing learners and in education, it is important to examine and evaluate the differences between
educators to adopt distance learning. Almost overnight, the education systems traditional and online learning. Understanding the differences between the two
scrambled to transition completely to online distance learning with little time education methods would clarify the effect of this sudden change from
to plan and no clarity of what will happen next. Teachers and students who are traditional to online learning and the impact it has left on learners.
used to interactive courses have begun to revise the curriculum and their learn-
ing methods to adapt to the new remote conditions. The spiraling crisis has
overwhelmed every aspect of the teaching world. After a pneumonia epidemic
2 Traditional Learning
without apparent cause, a novel strain of the Coronavirus – SARS-Cov-2 – was Learning is recognized as an active progression that cannot be obtained but
∗
Advised by: Colin Quirk must be built by the learner. Traditional learning consists of four critical el-
ements: an educator, or a knowledgeable person to share their proficiency; a
conducive setting like a classroom or lab; materials such as textbooks, videos,
1 2
diagrams, and charts; and motivation to personalize learning. The purpose of through experiences and the trial and error technique refers to experiential learn-
an educator is to facilitate the learning process by encouraging and providing ing. According to American educationalist theorist, David Kolb, there are four
support to a student to actively create knowledge, building upon ideas that the stages of experiential learning which include concrete learning, reflective obser-
learner already understands, for them to achieve their highest potential [D.08]. vation, abstract conceptualization, and active experimentation [K.09a]. Under
Domin states, “Knowledge cannot be transferred from one person to another; Kolb’s theory, an effective learner must go through all the following stages to
it must be actively constructed by the learner through interactions with the thoroughly be knowledgeable about a topic. For instance, a person cannot learn
environment” [S.99]. A designated environment limits distraction and focuses by memorizing given information. Instead, they need to take the extra effort
the student on active and interactive learning. Reference models and hands-on to become involved in the experience. A participant needs to reflect and recog-
experiences are tools used in traditional learning that motivate a student to take nize any common patterns or themes to help them with new experiences and to
initiative to learn. Learning in a classroom setting with strong student-teacher develop a more concrete understanding to apply in the future. Lastly, it is im-
and student-student relationships is an important component of traditional ed- portant to improvise, take risks, and experiment with new evolving theories and
ucation. Face-to-face instruction allows educators to assess the knowledge and circumstances to discover new methods of improvement. This technique allows
the progress of a learner to tailor teaching towards the development of the stu- learners to ‘learn by doing’ and apply their knowledge to future situations.
dents. They teach their expertise based on modules and concepts that encourage
student participation with the use of interactive tools such as group work, class
discussions, debates, peer critiques, and more [C.06]. Group work allows stu- 3 Online Learning
dents to discuss ideas amongst each other, developing deeper and more polished
understandings of subject matters. The promotion of active learning and col- The Coronavirus epidemic has become one of the largest challenges the educa-
laboration amongst educators and other peers to expand and build upon their tional system has ever had to face. Students from all over the globe are forced
to adapt to new circumstances in all areas of their lives, as governments have
prior knowledge. Learning is most effective when treated as a collaboration
authorized the ceasing of face-to-face education. As everyone obliged to shelter
between peers rather than individual competition. Social interaction frequently
in place rules, education platforms have changed, transitioning to an online, or
improves student participation and allows for the sharing of ideas, sharpening
distance learning model [J.20]. Within a short period, students were required to
students’ understanding. The zone of proximal development refers to knowl-
switch from traditional education to online and virtual learning. Online learn-
edge that is unobtainable for an individual independently but can be learned
ing is definitionally education that is executed through technology without the
with guidance from another person. Active discussions and asking questions
physical attendance of classes, lectures, or seminars. Remote education can be
allow the instructor to evaluate and determine what the learners know and do
further categorized as synchronous and asynchronous e-learning. Synchronous
not know and their average zone of proximal development [L05]. In traditional
e-learning refers to remote instruction where teachers and students communi-
education, this assessment is especially helpful for a teacher to scrutinize stu-
cate and collaborate electronically at the same time. Asynchronous e-learning
dents’ performance and provide suitable commentary and criticism. Traditional
is academic instruction without communication between the students and the
learning improves critical thinking through the sociability of an in-person class-
teacher. This can occur via online forums, interactive school courses, emails, vir-
room environment. Discussions and debates, for example, allow a student to
tual telecommunications, and so forth. Online education has not been reported
hear different perspectives and gain more understanding about a subject mat-
as an effective medium for learning as it lacks proper structure and necessary
ter. Critical thinking is an advanced way of thinking to analyze and assess a
social interaction. Although there is limited research regarding the quality of
judgment, centering on comprehension, analysis of various ideas and perspec-
online learning, there are multiple adverse experiences of remote learning which
tives, and problem-solving [H.02]. It is an important skill that inspires a broader
are based on poor online courses and pedagogy [J.05]. In comparison to tra-
mindset and analytical thinking. An individual should be able to recognize and
ditional learning, distance learning does not provide the needed discipline for
approach a problem, raise questions that challenge ideas that are simply in-
successful and efficient learning. Through traditional learning, an educator
formed to them, and develop creative solutions. Students who are taught in
a classroom structured environment as a unit, learn concepts through a “lec- can easily identify a student’s attentiveness and assess their teaching methods
ture and questioning” method. This technique emphasizes critical thinking as to ensure constant growth in a learner’s progress. However, in remote
it stimulates interactions between a learner, their peers, and an educator [C.06]. learning, it is difficult to keep an eye on a large number of students in a class
The mere “possession” of knowledge is not enough for critical thinking but on a screen. There are a multitude of distractions, interrupting the class flow
requires further motivation and a desire to learn [E.03]. Active learning can and creating challenges to teach the class. Distance learning does not allow
also be utilized outside of classroom walls and structured exercises. There are instructors to maintain the same student-teacher relationships as it was in the
multiple learning opportunities available for students such as simulations, in- classroom environment. As many institutes have converted to online
ternships, externships, autonomous studies, and hands-on programs. Learning education, it is a harsh reality that many students do not have access to the
technology and materials
3 4
needed. Students may not have a conducive environment or an appropriate 4 Effects of Traditional versus Online Learning
device needed to support their learning remotely. Underprivileged children, in
particular, discussed later in the paper, may not have access to the same level 4.1 Stress
of technology at home – whether it is the latest model of the laptop or high-
Stress can be defined as “when the perceived pressure exceeds your perceived
speed internet. In contrast, traditional, in-person schools are settings dedicated
ability to cope” [PSCCLT03]. In 2009, research done by Sulaiman, Hassan,
to the student’s education and give them equal access to all materials, while
Sapien, and Abdullah demonstrated that a majority of students experience
online learning resources may not be uniformly available to every individual.
some form of stress to some degree. In this study, the factors were subdi-
Whether the signal is lost, content is missed, or electronics are not available,
vided by examining potential academic stressors including schoolwork, grades,
the lack of accessibility to such materials is one of the reasons virtual learning
and overall performance in school; and personal stressors that may exist in
is not effective and cannot be a replacement for traditional learning. Learn-
a learner’s life such as extra-curricular involvement, self-esteem, and relation-
ing is not just an intellectual process, but a social activity, requiring in-person
ships [K.09b]. Stress, usually considered as a negative connotation, can bring
contact amongst other materials. Through the means of traditional learning,
an aspect of growth to certain conditions and individuals. Students deal with
students can interact with other individuals including their teachers, faculty,
positive and negative stress on a daily basis. Positive stress, otherwise known
and their peers. However, while engaging in online education, the learner is
as eustress, is invigorating as it is associated with conditions that provide chal-
not directly interacting with the educator and their classmates, making com-
lenges and opportunities for growth. Those who stress positively have an open
munication difficult and impersonal. Missing the day-to-day connections with
mindset perspective, correlated with more success in learning and accomplish-
a fellow student or teacher and the subsequent need to be self-motivated to
ing their goals and ambitions. They often view obstacles as experiments to test
push through schooling can prompt feelings of isolation. Discussions, for exam-
themselves and find methods of improvement. In contrast, negative stress also
ple, are a necessary component of social interaction between teachers, students,
referred to as distress, is correlated with threatening situations and the feeling
and classmates. This social activity demands active participation and allows
of helplessness [A.84]. Students will often feel powerless or lost, adversely af-
for students to engage in active cognitive processing. They mandate students
fecting their performance and prohibiting themselves to achieve their success.
to articulate their knowledge, encouraging them to engage, contextualize, and
Stress has a significant impact on a student’s academic performance. Of all the
apply what they know. According to the constructivist theory, learners require
factors identified by student reports, stress was identified as the leading cause
the ability to engage in opportunities to create meaning for themselves [S15].
that was negatively affecting an individual’s performance such as receiving a
Although discussions can be executed online, the inability to comprehend body
lower grade, not completing, or dropping out of a course [Ass10]. Reports illus-
language and chaos becomes an uninviting and hostile environment. Contrary
trate trends that show more students are experiencing stress from this sudden
to passive activities such as reading texts and listening to lectures, discussions
change, which must be acted upon for the overall safety and wellbeing of all
necessitate learners to analyze and decipher what they have learned in their own
learners. Learners who are incapable of coping with stress develop tension, un-
words. Discussions excellently allocate students to hear different perspectives
easiness, and anxiety. Stress this severe can lead to detrimental consequences
and compare and contrast ideas to enhance and better develop understanding
on a learner’s physical and mental health, affecting their performance
and ideas of their own. Education and learning go beyond academics and lesson
[K.09b]. Multiple studies elucidate the importance of social support in
plans; it also includes discipline, manners, morals, and social interactions. These
maintaining an individual’s physical and psychological health. There are three
traits are difficult to teach remotely. The authors of “The Science of Learning
different forms of coping techniques identified which include: problem-
and the Art of Teaching,” Jerome Feldman and Doug McPhee, proclaim that
focused coping, which correlates to the concept of identifying and confronting
kinesthetic learners are most successful when they are involved in an interac-
the source to relieve stress. Emotional-focused, or social coping refers to
tive activity that further promotes their learning [J.08]. Knowledge is easier
handling one’s emotional response to a stressor, and third, avoidant coping,
to comprehend if it is being constructed by the learner who has undergone the
which refers to the avoiding of the stressor as much as possible
intellectual process of reflection and analytical thinking. Students can retain
[MCLAAB12]. Social support is a way of emotional coping where there are
more information at a quicker pace as they participate in a lab, presentation,
social connections available to an individual for any form of support or
skit, field trip, or other activities. They can take the information they receive
reassurance. Four types of social support carry their individual benefits:
and can evaluate, experiment, and modify their ideas while participating in such
Emotional social support, the encouragement of one’s self-worth.
comprehensive processes that are available through traditional education [R.05].
Informational social support, relating to the sharing of advice and guidance to
Distance learning lacks the engaging and active experimentation process that
someone who may be undergoing a stressor. Tangible social support
is necessary for an effective learner to successfully acquire knowledge about a
includes the sharing of resources to relieve a stressor, and lastly, belonging
topic.
social support, which refers to the act of offering inclusion [OFJDCDEMI07].
Furthermore, it appears that social support of all types can help to construct
5 6
and strengthen the resistance to stress. Coping mechanisms of stress that in- a challenge for passion and improvement. Competition amongst peers can in-
volve social support heavily depend on communal interactions. With the recent stigate stress and drive motivation – both intrinsic and extrinsic – to do better
transition from traditional to online learning, social support coping processes and have the satisfaction of completing the activity, or to achieve some sort of
can not be as effective to relieve the stress of learners as before. Through reward [F.11]. Eisenberg and Thompson experimented with two similar condi-
traditional learning, students would frequently use their leisure time to reach tions to determine how competition affects the performance of improvisers.
out and interact with other classmates for emotional social support to relieve In the competition condition, participants were tasked to formulate an
some of their stress. With the current absence of in-person communication in improvised musical piece that would be blindly judged to determine the “best
distance learning, students may find it more difficult to reach out and interact improviser.” In the condition with no stimulated competition, participants
with their peers, limiting the availability and value of emotional social were told that the experimenters were interested in how people improvise.
support. In traditional learning, students can easily connect with teachers and The study illus- trated that competition, a combination of motivation and
counselors for the proper guidance and information social support needed to eustress, results in higher creativity and an overall boost in performance
fully comprehend the knowledge they are learning. However, there are more [F.11]. A learner’s educational development and motivation are influenced by
difficulties and obstacles regarding communication between students and “how” rather than “what” they are taught. Every individual has different
educators to ask questions, develop a thorough understanding of the motivations from comfortable to risk-taking environments. There are
curriculum, and build personal relationships with teachers in a remote- different ways a student can be inspired such as hands-on and interactive
learning environment. Tangible social support includes all the information a activities that challenge and test a person’s abilities. A traditional learning
classmate can receive from their peers about how to effectively deal with a environment provides a conducive setting for students to dedicate time and
course. In traditional learning, students can gain lots of tangible social space for their education. On the other hand, remote learning can take place
support from the sharing of resources available such as class or lecture notes. in any setting, which can disrupt the line between personal and professional
On the contrary, the available materials through the distance learning model environments. The task to keep a student motivated and actively engaged is a
are limited to a student. Outside of academics, social support refers to the challenge amongst all ages. Online learning courses, however, present
sense of support an individual can receive from social groups and teams that themselves with more concerns as new difficulties arise. To be successful,
center on teamwork and involvement. Extracurricular activities that occur students need to be disciplined, empowered, and self-regulated. Without the
with traditional learning, such as team sports, clubs, volunteer work, and many elements of traditional learning such as face-to-face contact, educators
charities allow students to participate and create a feeling of belonging to a are not able to detect any nonverbal clues from learners that signal their
certain group with similar interests or situations. Distance learning, social dis- disengagement to the course.
tancing regulations, and shelter in place isolation rules prohibit and hurdle the
group-related participation activities that provide belonging social support to 4.3 Secondary Effects: Relations
an individual.
An individual learns through social interactions, acquiring knowledge, and per-
4.2 Motivation sonality socially. According to the social constructivism theory, founded by
Russian psychologist Lev Vygotsky, learners have more growth in their develop-
Motivation is a powerful tool that disciplines and compels an individual to move ment when they incorporate the experiences, knowledge, and opinions of others
forward with a goal or objective. It refers to the “why” of our actions and be- to improve their learning [KCP09]. The evolution of individuality – personal to
haviors which are usually goal-oriented. Motivation can arise from external everyone – progresses with the agents of personality change: social relationships.
(extrinsic) or internal (intrinsic) factors. Extrinsic motivation is when individ- Social collaborations act as building blocks to one’s character in their personal-
uals are inspired to behave or participate in activities because of exogenous ity development. Personality can identify as an individuals’ psychological and
components. This example of motivation can come from another person such behavioral pattern, such as the perceptions and emotional feelings that are dif-
as an educator and or friend or an incentive to seek a certain reward or escape ferent from each person. Evolving from social situations, these characteristics,
punishment. A learner’s reasoning to be involved in an activity is an expectation and traits, are responsible for shaping an individual’s personality. Through the
to receive something in return such as a grade or praise or to avoid something simple means of communication, relations with others influence a person’s char-
negative including timeouts or a reduction in grades. Intrinsic motivation is the acter, allowing themselves to understand and develop their likes and dislikes,
engagement of activity for reasons such as passion and self-growth. The activity morals and ethics, and experiences and skills. Through the medium of online
is a reward in itself and is performed for the individual rather than an aspira- learning, social interactions which are a key component in the development of
tion for an external incentive. A learner would have a perspective of a growth personality, individuality, and learning is limited to everyone. Traditional learn-
mindset, partaking in an activity for their interest and further development of ing provides a conducive environment for social interactions. In a traditional
their construct of knowledge. They would view things with eustress, taking on learning environment, students connect in multiple different settings
7 8
including engaging activities in a teaching space (group projects, debates, and to attend their online schooling. Multiple Chinese news reports have stated
discussions) and outside the classroom (extracurricular activities and team that children have walked and climbed for hours at a time to mountaintops for
sports). Commu- nication between learners allows them to progress and be an adequate signal for remote classes [R.20]. An important element previously
exposed to new perspectives, expanding their knowledge. These connections identified for effective learning was the need for a conducive learning
influence their views on many topics that concede a person to formulate their environment. The transformation to online learning removes the
individuality [Nix13]. The ability to collaborate with peers is not an easy option accessibility of many pupil’s designated learning settings. Traditional learning
through technology screens in comparison to a same-setting classroom or offers classrooms and labs which are examples of encouraging environments
campuses. Beyond academics, traditional schooling impresses core social and that aid in the teaching and learning processes through hands-on activities
life skills that are crucial for success in the future. Social interactions improve between classmates. Remote learning requires pupils to attend their classes
and enhance the critical thinking and problem-solving abilities of learners as from their homes, which may not shelter the same designated settings for
they are subjected to the different opinions and assessments of their peers everyone. Some students, for instance, may have too many distractions or may
they collaborate with [Nix13]. Communication and public speaking, for have the added responsibilities of a sibling, pet, or elderly, taking away from
example, are common fears and vital aptitudes that can help a student in their ability to concentrate on online learning. Online learning has further
the workplace. Traditional learning allows students to speak in large groups, disadvantaged underprivileged students of lower socioeconomic status.
practicing the art of public speaking, as students present and articulate their Learners from low SES households and communities are often challenged
thoughts and ideas in front of their educators and peers. In distance with the lack of the needed equipment and conducive space to attend distance
learning, speaking through virtual chats via computers does not teach public learning classes. Traditional learning, on the other hand, offers students equal
speaking. resources and a designated place for the education of all students.
5 Underprivileged Learners 6 Discussion

As the Coronavirus pandemic continues to spread, millions of learners across the The Coronavirus pandemic has radically shifted traditional learning mechanisms
world are affected by school closures. The abrupt compulsion to convert to to online learning. As the world tackles the sudden change to distance learn-
online learning has caused deep digital divides. Underprivileged students, ing, students are missing important experiences that supplement their learning
whose family, social, or economic conditions impair their learning, are already process. Attending remote classes in isolation at home confines a student, keep-
at a disadvantage and are most susceptible to these technological gaps. Often ing them from important social interactions that encourage learning. A nod in
these students do not have access to technology or an internet connection, the hallway, a wave in the cafeteria, or a smile in the locker room are a few
which along with an overly stimulating home makes it difficult to concentrate. examples of social gestures that help a student cope with stress. Social interac-
Re- search shows that learners from low socio-economic status families and tions appropriately address stress and motivate students, which is not remotely
societies progress slower in academic skills in comparison to those belonging accessible. Hands-on involvements such as field trips, lab work, and group as-
in higher SES classes [MPLF09]. For example, growing up in a low SES signments, are also examples of crucial elements in educational development.
community is associated with poor cognitive performance, linguistics, The absence of these important components negatively impacts the education
memory, and the overall understanding of subjects. The educational systems of students. The COVID-19 continues to reveal new layers of inequity within
in low socio-economic status areas frequently lack resources which adversely societies. An example of a vulnerable population with negative effects on their
affects the academic development and performance of the students [N.15]. learning are children growing up in underprivileged societies. The absence of
The addition of remote classes further challenges the underprivileged pupils. in-person schooling adds additional challenges to their education. Their inac-
Many learners from low socio-economic backgrounds are not able to attend cessibility of a designated quiet space and other fundamental equipment such
online classes for an array of reasons such as sharing or not having access to a as a device and internet connectivity place these children at a major disadvan-
laptop or device to be present at a class [L.20]. According to data collected tage. As more than half a year has passed since school closures, the new school
over four weeks, ten percent of low SES children have little to no access to year has circled, leaving many in disagreement regarding the opening of schools.
the equipment needed for online learning [Bur20] More than twenty-five As the media covers one of the most controversial debates, it is important to
percent of children who are experiencing food insecurity cannot access the remember the decision to re-open schools should not be driven by the pressure
internet to attend distant learning classes [Bur20]. The sudden change to of politics but the evidence of science. As the current situation lingers and is
distance learning has forced students to take the extra steps to participate in becoming a new “normal” of learning, there are measures that can be taken
classes through the means of technology, which may not be available to to better improve upon the current conditions of education and fill the gaps
many learners. Motivated to learn, many students are taking extreme
measures to obtain access to the needed materials
10
9
between traditional and online learning. For instance, it is important to make [fDCP20] Centers for Disease Control and Prevention. Coronavirus
adjustments to ensure equal resources and access to all learners. Additional disease 2019 control and prevention. Centers for Disease
funding from the government can be provided for schools to be more resilient Control and Prevention Advisory Feb 11, 2020, 2020.
and be able to offer electronic materials such as laptops and hotspots to all
students who do not have access to these materials. Public libraries can open [H.02] Astleitner H. Teaching critical thinking online. Journal of
with appropriate safety precautions to accommodate for a conducive setting for instructional psychology, 2002.
learners to be able to access a designated space to facilitate their learning. Ed- [J.05] Rovai A. P. Wighting M. J. & Liu J. School climate. Quar-
ucators can grant office hours to students with special needs or highly terly Review of Distance Education, 2005.
stimulus households for any additional help or support they may need. The
potential solutions to this ongoing educational crisis will require more funding, [J.08] Tranquillo J. Ac 2008-599: Kinesthetic learning in the class-
new poli- cies, innovative lesson plans, new etiquettes, new attitudes, and room. age, 2008.
much more, in which the communities and government have a crucial role to
[J.20] Daniel S. J. Education and the covid-19 pandemic. Prospects,
play.
2020.
[K.09a] Abdulwahed M.& Nagy Z. K. Applying kolb’s experiential
References learning cycle for laboratory education. Journal of engineer-
[A.84] Whitman N. A. Student stress: Effects and solutions. ASHE- ing education, 2009.
ERIC Higher Education Research Report, 1984. [K.09b] Sulaiman T. Hassan A. Sapian V. M. & Abdullah S. K. The
[ARTON] AUTHOR OF ARTICLE. Title of article. NAME OF JOUR- level of stress among students in urban and rural secondary
NAL, YEAR OF PUBLICATION. schools in malaysia. European journal of social sciences,
2009.
[Ass10] American College Health Association. Reference group exec-
utive summary fall 2010. American College Health Associa- [KCP09] K. C.. Kalina C. & Powell. Cognitive and social construc-
tion Executive Summary Fall 2010, 2010. tivism: Developing tools for an effective classroom. Educa-
tion, 2009.
[Bur20] US Census Bureau. U.s. census bureau releases household
pulse survey results. The United States Census Bureau Press [L05] Limbach B. J. & Waugh W. L. Questioning the lecture for-
Release 2020, 2020. mat. The NEA Higher Education Journal, 2005.
[C.06] Sungur S. & Tekkaya C. Effects of problem-based learning [L.20] Iivari N. Sharma S. & Ventä-Olkkonen L. Digital transforma-
and traditional instruction on self-regulated learning. The tion of everyday life–how covid-19 pandemic transformed the
journal of educational research, 99, 2006. basic education of the young generation and why information
management research should care? International Journal of
[CDC20] CDC. Severe acute respiratory syndrome coronavirus 2 from Information Management,, 2020.
patient with coronavirus disease, united states. Emerging
Infectious Diseases journal, 2020. [MCLAAB12] R. D. MacCann C. Lipnevich A. A. Burrus, J. & Roberts.
The best years of our lives? coping with stress predicts
[D.08] Wood K. C. Smith H. & Grossniklaus D. Piaget’s stages. De- school grades, life satisfaction, and feelings about high school.
partment of Educational Psychology and Instructional Tech- Learning and Individual Differences, 2012.
nology, University of Georgia, 2008.
[MPLF09] G. Hillemeier M. M. & Maczuga S. Morgan P. L. Farkas. Risk
[E.03] Walker S. E. Active learning strategies to promote critical factors for learning-related behavior problems at 24 months
thinking. Journal of athletic training, 2003. of age: Population-based estimates. Journal of abnormal
child psychology, 2009.
[F.11] Eisenberg J.& Thompson W. F. The effects of competition
on improvisers’ motivation, stress, and creative performance.
Creativity Research Journal, 2011.
11 12
[N.15] Barbarin O. A. & Aikens N. Overcoming the educational dis-
advantages of poor children: How much do teacher prepara-
tion, workload, and expectations matter. American Journal
of Orthopsychiatry, 2015.
Melanoma Diagnosis using Convolutional Neural
[Nix13] Hurst B Wallace R. Nixon. The impact of social interaction
on student learning. Reading Horizons: A Journal of Literacy Networks
and Language Arts, 2013.
∗
Ansh Chaurasia
[OFJDCDEMI07] S. Ozbay F. Johnson D. C. Dimoulas E. Morgan III, C. A.
Charney D. & Southwick. Social support and resilience to February 2021
stress: from neurobiology to clinical practice. Psychiatry
(Edgmont), 2007.
[Org20] World Health Organization. Coronavirus disease 2019 (covid-
Abstract
19) situation report – 94. WHO Situation Reports April 23,
2020, 2020. Melanoma is one of the most lethal forms of skin cancer in the USA,
with an estimated 196,000 people being diagnosed by melanoma in 2020.
[Pro20] United Nations Development Programme(UNDP). Coron- Fortunately, melanoma can be much more straightforward to treat through
avirus disease covid-19 pandemic. UNDP Paper July 7 2020, skin excisions once it has been diagnosed, once again placing the brunt
2020. of the challenge on diagnosis. One of the primary techniques used by
dermatologists to diagnose melanoma is known as dermoscopic imaging,
[PSCCLT03] K. Palmer S. Cooper C. L. & Thomas. Creating a balance: which use high quality magnifying lenses to capture skin lesions in great
Managing stress. British Library Board, 2003. detail. We believe that we can take advantage of this prominent imaging
technique to create a Convolutional Neural Network that receives these
[R.05] Wolfe M. B. & Goldman S. R. Relations between adolescents’ dermoscopic images as input and outputs a binary classification - whether
text processing and reasoning. Cognition and Instruction, 23, the melanoma is malignant or benign. To do this, we took a 100 GB Kag-
2005. gle dataset from a Kaggle Competition and applied image transformations
upon it to rigorously train a machine learning ensemble of models. Fur-
[R.20] Zhong R. The coronavirus exposes education’s digital divide. thermore, we experimented with various machine learning architectures,
NY Times March 18, 2020. techniques, and metrics to come up with a machine learning model that
returns predictions with emphasis on accuracy and efficiency, boasting
[S.99] Domin D. S. A review of laboratory instruction styles. Jour- high AUROC and accuracy scores.
nal of chemical education, 1999.
[S15] Bada S. O. & Olusegun S. Constructivism learning theory: 1 Introduction
A paradigm for teaching and learning. Journal of Research
& Method in Education, 2015. A recent report by the World Health Organization (WHO) has included cancer
in the top 10 causes of death [WHO]. More alarmingly, data from the same
[UNE20] UNESCO. Education: From disruption to recovery. UN-
report also indicates the rate of patients diagnosed with cancer may double
ESCO Report September 1 2020, 2020.
[Aus18]. Cancer can be lethal to patients, but its effects can be mitigated if
detected and treated early [ski]. Therefore, it is worthwhile to invest time and
research to improve our ability to diagnose cancers as early as possible.
Melanoma is one of the most lethal forms of skin cancer. It occurs in cells
known as melanocytes, skin cells in the upper layer of skin. Melanocytes pro-
duce a pigment known as melanin to give the skin it’s color. However, when skin
is exposed to UV radiation, melanocytes produce more melanin than necessary,
∗ Advised by: Mr. Jeremy Irvin, Stanford University
13 14
causing skin damage. Melanoma occurs when UV radiation causes mutations in
these melanocytes, which leads to unrestrained cellular growth. Figure 1. gives
a visual difference between benign and malignant melanoma images. By the end
of 2020, about 196,060 people in the USA will be diagnosed with melanoma, and
of these more than 100,000 people are expected to be diagnosed with invasive
(penetrating the epidermis into the skin’s second layer, the dermis) melanoma
[noad],[ski]. About 6,850 patients suffering fatally from melanoma are likely to
have died in 2020 [noaa]. Unfortunately over the past 40 years, melanoma cases
have been steadily rising [noab]. Amidst this melonomic gloom, the good news
is that melanoma can be cured through excisions when detected and diagnosed
in its early stages [CKU+ 07],[CCB+ 11]. At present, the available detection
and diagnosis options for melanoma are visual inspection, clinical screening,
dermoscopic analysis, biopsy and histopathological examination of skin lesion.
Among all options, dermoscopy is the most popular imaging technique. Der-
moscopy refers to microscopic examination and evaluation of skin lesions. It is
typically done with every high quality magnifying lens and powerful illumina-
tion system (aka Dermatoscope [noac]). However, dermoscopic images are not
easy to interpret for diagnosis. Even with most experienced dermatologists, the
evaluation of dermoscopic images can be laborious and error prone [CIS+ 08],
[ACS+ 13]. The complex visual characteristics of skin lesions such as multi-
sizes, multi-shapes, fuzzy boundaries, and low contrast when compared to the Figure 1: Two sets of dermoscopic images. Images on the left column show
skin and noise presence such as skin hair, oils, air, and bubbles limit even an benign cases, while the images on the right show malignant cases
expert dermatologists’s sensitivity to less than 80% [VMHM08]. Figure 1. gives
a visual feel of some lesion images as classified by multiple expert dermatologists
into benign and malignant melanoma. networks have been known for being effectively utilized to diagnose cancers and
Aforesaid challenges especially motivate the machine learning community to diseases in the past [CKK18]. Some recent examples of using CNNs for medical-
design algorithms to automatically diagnose melanoma in dermoscopic images. imaging related tasks include diabetic retinopathy detection which performed
Computer-aided diagnosis (CAD) system automates interpretation of dermo- on levels comparable to ophthalmologists [GPC+ 16], chest X-Ray pathology
scopic images to diagnose melanoma. This helps in early and successful diag- detection in chest radiographs which matched the performance of radiologists
nosis of melanoma, thereby making the treatment effective and reducing the [BNG+ 19] , and knee abnormality detection in MRI scans [BRB+ 18]. Given the
mortality rate. Machine learning methods aim to ‘train’ models using labeled recent success of convolutional neural networks on a variety of medical imag-
data (dermoscopic image of benign and malignant melanoma) and then provide ing tasks, there is a significant opportunity for research on developing models
prognosis on the new dermoscopic images of patients. Recently, a subfield of ma- for other tasks [WFT+ 19]. One of the early works using CNN to classify der-
chine learning known as deep learning has shown success in automatic medical moscopic images into benign and malignant, with accuracy levels comparable
image interpretation comparable to the level of human specialists. Deep learn- to that of 21 board-certified dermatologists, came from Esteval et. al in 2017
ing focuses on the construction of “neural networks”, which comprises multiple [EKN+ 17]. In this research paper, I developed deep learning models to classify
layers of non-linear functions with coefficients/weights which are derived from dermoscopic images of skin lesions. The networks were trained and validated on
the training data. These weights are the result of the ‘training’ process and a large dataset of dermoscopic images that were labeled by dermatologists as be-
hold the knowledge of differentiation between benign and malignant melanoma. nign or malignant. I discovered that a ResNet model with 50 layers achieves an
Artificial Neural Networks (ANNs) were the first type of neural network to AUROC score of 0.78 on the validation set, while an EfficientNet model achieves
demonstrate success in medical imaging classification, including CadE/CadX a AUROC score of 0.90. The model leverages augmentations and ensembling of
to diagnose chest diseases given radiographs [QYSS18], diagnose general cancer multiple smaller models to achieve that high performance on the validation set.
given tumor and lymphatic node images [PMC16], and diagnose prostate cancer Finally, I interpreted the model predictions through the use of Class Activa-
given MRI images [DVBF+ 16]. Researchers have developed specialized neural tion mappings to highlight where in the picture the model considered a possibly
networks known as convolutional neural (CNN) networks which are designed malignant tumor existed.
to model the highly structured data present in images. Convolutional neural
15 16
2 Methods
Training Validation
2.1 Data Positive No (%) 473 (1.8%) 111 (1.6%)
Negative, No (%) 25,459 (98.2%) 7,083 (98.4%)
The dataset consisted of 33,126 dermoscopic images in total and was split into Mean age 45.0 45.0
a training set (25,932 images) to learn model parameters and a validation set % Female 48.9 % 45.9 %
(7,194 images) to compare models. As shown in Table 1., the training set con- Total 25,932 7,194
sisted of 473 positive (malignant) cases or 1.8 % of the training set. The 25,459
Table 1: Data statistics across the data splits.
images remaining were negative (benign) cases or 98.2 % of the training set.
The validation set consists of 7,194 images. As shown in Table 1, the valida-
tion set consisted of 111 positive (malignant) cases or 1.6 % of the validation
set. The 7,083 images remaining were negative (benign) cases or 98.4 % of the
validation set. Both datasets had the exact same approximate age of 45, with
the percent of female patients varying from 48.9% in the training set to 45.9% 2.2 Convolutional Neural Networks
in the validation set. All images in the training and validation are in JPEG
Convolutional neural networks are especially effective in machine learning due
format and were resized to 224 x 224 pixels.
to their ability to leverage the structured format of imagery. Convolutional
networks are composed of convolutional layers, which use a kernel and stride to
extract certain features from the image. Conv layers are followed by non-linear
Rectifier layers (aka ReLU in Figure 2.) which typically remaps the input to
a manageable ‘range’ (e.g. -1 to +1). Typical CNNs are combinations of the
three layers (convolution, rectifier, pooling). The final layer of a CNN is a fully-
connected layer to produce the final output of the network (classification label).
Certain CNN architectures are more effective for different types of tasks. To
mitigate overfitting, two candidate models were investigated in this work for
their relatively small neural network size - ResNet50 [HZRS15] and EfficientNet
[TL19]. A diagram of the ResNet50 architecture is shown below with Figure 2.
and the EfficientNet Architecture is shown in Figure 3.
Figure 2: Diagram of the ResNet50 Architecture
In essence, the ResNet50 architecture functions by stacking up sets of the

three layers 50 times [HZRS15]. The triplet consists of convolutional layers,
normalization / pooling layers, and rectifier layers. The convolutional layer
17 18
extracts relationships from the image, the pooling layer aggregates the extracted
features, the rectifier nonlinearities (ReLU) are applied to capture nonlinear
structures within the image. One aspect that differentiates ResNet from other
CNNs - is that it trains to learn the ‘residual signal’, which refers to the difference
of a triplet’s output with input of the previous triplet.
Figure 3: Diagram of EfficientNet Architecture Used

Figure 4: A Conceptual Diagram of how melanoma diagnosis is accomplished
The EfficientNet architecture is composed of layers similar to the ResNet50 in the model.
architecture but also efficiently balances the tradeoff between network depth
(number of layers), width (number of channels), and image resolution. It is 8.4
times smaller and 6.1 times faster than the best convolutional neural network 2.4 Evaluation Metrics
and achieves one of the highest accuracies on the ImageNet database, a large I used a variety of performance metrics to evaluate the deep learning models.
dataset of natural images which is the most commonly used benchmark for To evaluate the quality of the model probabilities, I computed the receiver op-
assessing convolutional neural network performance 24. erating characteristic (ROC) curve and precision recall curve (PR) . Then, to
compress the curves into more comprehensive metrics, the area from the ROC
2.3 Training Procedure and PR curves was computed to get AUROC and AUPRC. The model was pri-
marily evaluated using the AUROC score, but AUPRC was also measured to
A special training procedure was created to maximize the performance of the
provide a holistic evaluation of the models. To convert probabilities to binary
ResNet50 / EfficientNet models. In order to utilize the advantages of transfer
predictions and compute point metrics, I used the threshold which led to the
learning, both models had weights initialized from a network pre-trained on data
highest F1 score on the validation set. After the probabilities were converted
from ImageNet. Both models used a scheduler that reduced the learning rate
to binary predictions at the optimal threshold, I computed the precision, which
each time performance on the validation set plateaued, an Adam optimizer, and
measures how many accurate malignant diagnoses were made over all malignant
the binary cross entropy loss function. The Adam optimizer was utilized with
diagnoses the model predicted, recall, which measures how many accurate ma-
an initial learning rate of 0.001. The scheduler had a patience of 1, a threshold
lignant diagnosis were made over all test images which were malignant, F1 score
of 1 * 10− 5 to measure when the validation performance plateaued, and a factor which is the harmonic mean of precision and recall, and accuracy of the model.
of 0.2 to reduce the learning rate by. Each model was trained for a maximum
of 20 epochs (20 full passes over the training set) with a batch size of 32 and 4
workers (processors). The following augmentations were applied to each batch 2.5 Model Interpretation
during training: random horizontal and vertical flips with a probability of 50%, Once training and testing was complete, I sought to interpret the model pre-
random crops of the image, and random rotations of the image. After the dictions. In order to accomplish this, I used class activation maps (CAMs,
model completed iteration over the training dataset, the model was directly Figures 5/6.) to highlight the regions of the image which contribute most to the
tested on the validation set. The summary metric AUROC was computed for model’s prediction of melanoma. The class activation maps are computed using
each epoch and the 5 models with the highest AUROC score were saved. These the feature maps right before the global average pooling layer (GAP) is reached,
five models were used to generate an ensemble of models where the average of where feature map (1 per channel) is multiplied by the weight corresponding
the probabilities predicted by each model was used as the prediction for the to the final fully connected layer, and then summed to create the CAM. The
ensemble, as depicted in Figure 4. Finally, to limit training time and prevent CAM is a heatmap that overlays the image where temperature/color highlights
overfitting on the training set, early stopping was used which terminates training the regions of the image which contribute most to the model’s classification.
once the marginal gains from training were lower than a specific threshold. In Figures 5/6, the blue regions highlight areas which contribute more to the
prediction of melanoma, while purple highlights areas which contribute less. Us-
19 20
ing these CAMs, especially for false negative / positive diagnosis cases, helped 3 Results
me understand possible causes that were derailing the model from an accurate
prediction.
Point Metrics Summary Metrics
Experiment Precision Recall F1 Accuracy AUROC AUPRC
1 Pretrained 0.03 0.91 0.06 0.56 0.79 0.05
ResNet50,
with optimizer
Adam, no
early stopping,
and CrossEn-
tropyLoss loss
function
2 Pretrained 0.13 0.32 0.19 0.96 0.86 0.15
ResNet50,
with optimizer
Adam, no early
stopping and
Figure 5: Class activation maps (CAM) of the best model on the validation set. BinaryCrossEn-
The CAM on the left is an accurate diagnosis of benign melanoma, while the tropyWithLog-
CAM on the right shows a false diagnosis (the model incorrectly predicted there its (BCEL) loss
was melanoma). Areas that have higher hues of blue indicate the model has a function
higher confidence that melanoma exists in that part of the image, while areas 3 Pretrained 0.09 0.51 0.15 0.91 0.87 0.11
that are more purple indicate lower confidence. ResNet50 with
optimizer Adam
and scheduler,
no early stop-
ping, and BCEL
loss
4 Pretrained 0.07 0.71 0.13 0.85 0.88 0.14
ResNet50, with
optimizer Adam
and scheduler,
early stopping,
and BCEL loss
5 Pretrained Ef- 0.14 0.14 0.04 0.89 0.54 0.02
ficientNet with
optimizer Adam
and scheduler,
Figure 6: Class activation maps (CAMs) of the best model on the validation set. early stopping,
The CAM on the left is an accurate diagnosis of malignant melanoma, while the and BCEL loss
CAM on the right shows a false diagnosis (the model incorrectly predicted there
wasn’t melanoma). Areas that have higher hues of blue indicate the model has
a higher confidence that melanoma exists in that part of the image, while areas
that are more purple indicate lower confidence.
21 22
was replaced with an EfficientNet model. This caused the AUROC score to
6 Pretrained Ef- 0.21 0.20 0.20 0.98 0.87 0.13 plummet. In Experiment 6, 5 EfficientNet models were put into an ensemble
ficientNet with where the predictions would come as the arithmetic mean of the 5 predictions of
optimizer Adam the individual models. This increased AUROC from 0.54 to 0.87. The intention
and scheduler, for Experiment 7 was to train the model more rigorously by augmenting the
early stopping, training set images. Once the bug was fixed, the model outputted the highest
BCEL loss, and and final AUROC score at Experiment 8 - 0.9.
ensembling
7 Pretrained Ef- 0.42 0.16 0.23 0.98 0.57 0.08
ficientNet with 4 Discussion
optimizer Adam
and scheduler,
early stopping
with increased
threshold,
BCEL loss and
ensembling and
data augmen-
tation (Error
in generating
predictions)
8 Pretrained Ef- 0.1 0.66 0.28 0.91 0.90 0.17
ficientNet with
optimizer Adam
and scheduler,
early stopping
with high-
est threshold,
BCEL loss and Figure 7: ROC curve of best model of the validation set.
ensembling
Table 2: Performance metrics of the experiments on the validation In this work, I developed convolutional neural networks for the detection of
set. melanoma in dermoscopic images. Two different models, ResNet50 and Ef-
ficientNet, were investigated together with a variety of training procedures.
Through each experiment, one important variable was toggled, such as model
type, loss function, use of scheduler, early stopping, data augmentation, and en-
sembling. Overall, the key methods used for the greatest gains in AUROC came
I experimented with a variety of models and training procedures in order from the use of early stopping and ensembling of the models. Apart from the
to investigate their impact on performance, primarily in terms of AUROC.. model improvement methods, other techniques such as CAMs / graphs were
Starting with a ResNet50 model and Experiment 1 metrics as a baseline, the utilized to figure out optimal values for certain variables and to analyze the
loss function was changed to BinaryCrossEntropyWithLogits (BCEL) due to its models.
specialty in binary classification, which led to an increase inAUROC score from Major implications of our work lie mainly in the use of multiple techniques
0.79 to 0.86 (Experiment 2). Next, a scheduler was added in order to anneal applied, and the gain of each technique. The set of techniques and their re-
the learning rate (Experiment 3). Although the AUROC score does increase spective gains can be used by other researchers in the medical field to prioritize
from Experiment 2 to 3, the gain is relatively small. In Experiment 4, early which techniques to use to maximize their own models’ AUROC score. A second
stopping was added to prevent overfitting on the training set which also led to implication lies in the model’s AUROC score. As the dermatologist accuracy
minimal performance gains. In Experiment 5, the ResNet50 model being used of 75% [NFKA20] has been surpassed by our model’s 91% accuracy and 90%
23 24
References
[ACS+ 13] Qaisar Abbas, M E Celebi, Carmen Serrano, Irene Fondón Garcı́a,
and Guangzhi Ma. Pattern classification of dermoscopy images: A
perceptually uniform model. Pattern Recognit., 46(1):86–97, Jan-
uary 2013.
[Aus18] Australian Institute of Health and Welfare. Cancer in australia:
Actual incidence data from 1982 to 2013 and mortality data from
1982 to 2014 with projections to 2017. Asia Pac. J. Clin. Oncol.,
14(1):5–15, February 2018.
[BNG+ 19] Ivo M Baltruschat, Hannes Nickisch, Michael Grass, Tobias Knopp,
and Axel Saalbach. Comparison of deep learning approaches for
Multi-Label chest X-Ray classification. Sci. Rep., 9(1):6381, April
2019.
[BRB+ 18] Nicholas Bien, Pranav Rajpurkar, Robyn L Ball, Jeremy Irvin, Alli-
Figure 8: PR curve of best model of the validation set. son Park, Erik Jones, Michael Bereket, Bhavik N Patel, Kristen W
Yeom, Katie Shpanskaya, Safwan Halabi, Evan Zucker, Gary Fan-
ton, Derek F Amanatullah, Christopher F Beaulieu, Geoffrey M
ROC AUROC, the model proposed has the potential to match or exceed human
Riley, Russell J Stewart, Francis G Blankenberg, David B Larson,
performance, implying it may be viable to assist humans in the detection of
Ricky H Jones, Curtis P Langlotz, Andrew Y Ng, and Matthew P
melanoma. This work also has important limitations which should be consid-
Lungren. Deep-learning-assisted diagnosis for knee magnetic reso-
ered. First, data only came from 6 hospitals (Hospital Clı́nic de Barcelona, Med-
nance imaging: Development and retrospective validation of MR-
ical University of Vienna, Memorial Sloan Kettering Cancer Center, Melanoma
Net. PLoS Med., 15(11):e1002699, November 2018.
Institute Australia, The University of Queensland, and the University of Athens
Medical School), so the model may not be representative of certain populations [CCB+ 11] Germán Capdehourat, Andrés Corez, Anabella Bazzano, Rodrigo
[noae]. As a result, the model will likely be less generalizable and over-represent Alonso, and Pablo Musé. Toward a combined tool to assist derma-
the populations from where the dataset originated. Another limitation was the tologists in melanoma detection from dermoscopic images of pig-
oversimplification of the task. In the case melanoma is classified, the sever- mented skin lesions. Pattern Recognit. Lett., 32(16):2187–2196, De-
ity must also be determined. One metric to determine this is called the Clark cember 2011.
Scale, which ranks detected melanoma into 5 levels [Mel]. However in this work,
melanoma was only classified based on existence, ignoring many nuanced details [CIS+ 08] M Emre Celebi, Hitoshi Iyatomi, William V Stoecker, Randy H
that are important to the diagnosis of cancer. Finally, skin lesions in different Moss, Harold S Rabinovitz, Giuseppe Argenziano, and H Peter
parts of the body may be treated differently. The models presented in this work Soyer. Automatic detection of blue-white veil and related structures
do not utilize information about where the lesions originated in the body, which in dermoscopy images. Comput. Med. Imaging Graph., 32(8):670–
may be important information for classification. All such known limitations 677, December 2008.
could be addressed with help of availability of a variety of data which cov- [CKK18] S Charan, M J Khan, and K Khurshid. Breast cancer detection in
ers different geographies, races, regions of the body, and severity levels of the mammograms using convolutional neural network. In 2018 Inter-
melanoma. In future research, we plan to explore other model architectures, national Conference on Computing, Mathematics and Engineering
and versions of ResNet and EfficientNet which are available in PyTorch. Using Technologies (iCoMET), pages 1–5, March 2018.
K-Fold cross validation instead of saving the top 5 models would also be consid-
ered in order to improve the heterogeneity of models in the ensemble. Finally, [CKU+ 07] M Emre Celebi, Hassan A Kingravi, Bakhtiyar Uddin, Hitoshi Iy-
generating custom augmentations such as inserting obstructions like hairs into atomi, Y Alp Aslandogan, William V Stoecker, and Randy H Moss.
the images for the training set could also improve the robustness of the model. A methodological approach to the classification of dermoscopy im-
ages. Comput. Med. Imaging Graph., 31(6):362–373, September
2007.
25 26
[DVBF+ 16] Pieter J L De Visschere, Alberto Briganti, Jurgen J Fütterer, Pirus [PMC16] Seyedmehdi Payabvash, Kaan Meric, and Zuzan Cayci. Differen-
Ghadjar, Hendrik Isbarn, Christophe Massard, Piet Ost, Prasanna tiation of benign from malignant cervical lymph nodes in patients
Sooriakumaran, Cristian I Surcel, Massimo Valerio, Roderick C N with head and neck cancer using PET/CT imaging, 2016.
van den Bergh, Guillaume Ploussard, Gianluca Giannarini, and
Geert M Villeirs. Role of multiparametric magnetic resonance [QYSS18] Chunli Qin, Demin Yao, Yonghong Shi, and Zhijian Song.
imaging in early detection of prostate cancer. Insights Imaging, Computer-aided detection in chest radiography based on artificial
7(2):205–214, April 2016. intelligence: a survey. Biomed. Eng. Online, 17(1):113, August
2018.
[EKN+ 17] Andre Esteva, Brett Kuprel, Roberto A Novoa, Justin Ko, Susan M
Swetter, Helen M Blau, and Sebastian Thrun. Dermatologist-level [ski] skincancer.org. Skin cancer facts & statistics.
classification of skin cancer with deep neural networks. Nature, https://www.skincancer.org/skin-cancer-information/skin-cancer-
542(7639):115–118, February 2017. facts/. Accessed: 2020-10-21.
[GPC+ 16] Varun Gulshan, Lily Peng, Marc Coram, Martin C Stumpe, Derek [TL19] Mingxing Tan and Quoc V Le. EfficientNet: Rethinking model
Wu, Arunachalam Narayanaswamy, Subhashini Venugopalan, Ka- scaling for convolutional neural networks. May 2019.
sumi Widner, Tom Madams, Jorge Cuadros, Ramasamy Kim, Ra- [VMHM08] M E Vestergaard, P Macaskill, P E Holt, and S W Menzies. Der-
jiv Raman, Philip C Nelson, Jessica L Mega, and Dale R Webster. moscopy compared with naked eye examination for the diagnosis
Development and validation of a deep learning algorithm for detec- of primary melanoma: a meta-analysis of studies performed in a
tion of diabetic retinopathy in retinal fundus photographs. JAMA, clinical setting, 2008.
316(22):2402–2410, December 2016.
[WFT+ 19] Julia K Winkler, Christine Fink, Ferdinand Toberer, Alexander
[HZRS15] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep Enk, Teresa Deinlein, Rainer Hofmann-Wellenhof, Luc Thomas,
residual learning for image recognition. December 2015. Aimilios Lallas, Andreas Blum, Wilhelm Stolz, and Others. Asso-
[Mel] Breslow depth and clark level. ciation between surgical skin markings in dermoscopic images and
https://www.curemelanoma.org/about-melanoma/melanoma- diagnostic performance of a deep learning convolutional neural net-
staging/breslow-depth-and-clark-level/. Accessed: 2020-10-14. work for melanoma recognition. JAMA Dermatol., 155(10):1135–
1141, 2019.
[NFKA20] A Naeem, M S Farooq, A Khelifi, and A Abid. Malignant melanoma
classification using deep learning: Datasets, performance mea- [WHO] WHO. Cancer. https://www.who.int/news-room/fact-
surements, challenges and opportunities. IEEE Access, 8:110575– sheets/detail/cancer. Accessed: 2020-10-21.
110597, 2020.
[noaa] American cancer society. https://cancerstatisticscenter.cancer.org/#!/cancer-

site/Melanoma%20o%20the%20skin. Accessed: 2020-10-21.
[noab] American cancer society. https://cancerstatisticscenter.cancer.org/cancer-
site/Melanoma%20of%20the%20skin/nWHbpgnv. Accessed:
2020-10-21.
[noac] Dermoscopy. https://dermnetnz.org/topics/dermoscopy/. Ac-

cessed: 2020-10-21.
[noad] Melanoma statistics. https://www.curemelanoma.org/about-

melanoma/melanoma-statistics-2/. Accessed: 2020-10-12.
[noae] SIIM-ISIC melanoma classification.

https://www.kaggle.com/c/siim-isic-melanoma-
classification/data. Accessed: 2020-10-14.
27 28
The field of behavioural economics revolves around certain heuristics and
how they affect behaviour. The term heuristic is commonly defined as a cog-
nitive shortcut that simplifies decisions, especially under conditions of uncer-
tainty.2 As stated by Daniel Kahneman, heuristics represent a process of sub-
Behavioural Economics and Education: stituting a difficult question with an easier one.3 They can also lead to cognitive
Exploring possible methods by which heuristics biases. Several different heuristics exist and impact various sectors of human
life.
could be used to improve the performance of Education is fundamental to raising an independent and successful future
generation. This field is the foundation of the future, and optimising it for
students in the International Baccalaureate maximum benefit is essential. Thus, this research paper considers the impact
of heuristics on education and proposes methods by which heuristics can be
Diploma Program (IBDP) employed by schools, teachers, and students, to help improve the learning of
students and consequently improve student performance in examinations. The
∗
paper looks at the impact of heuristics and their application specifically on stu-
Arzoo Usgaonkar dents of the International Baccalaureate Diploma Programme (IBDP) but its
proposals could be applied to other curricula as well, albeit with some modifi-
April 2, 2021 cation. This paper considers the following heuristics:
1. The status quo bias
2. The conformity bias
3. The present bias
Abstract
This research report examines the effect of three heuristics – the sta-
tus quo effect, the conformity bias, and the present bias – on the per- 2 Context
formance of IB Diploma Program students, globally. It looks at several
previously conducted studies and relates their findings to teaching and The International Baccalaureate Diploma Programme (IBDP)4 is an education
learning methods of IBDP teachers and students, respectively. The pa- curriculum that strives to be challenging and aims to lead to the holistic de-
per considers several, possibly unnoticeable, short-comings of the current velopment of its students. It hopes that its students become inquirers, knowl-
education system, explains how heuristics may be employed to improve edgeable, thinkers, communicators, principled, open minded, caring, risk-takers,
the formal education system, and proposes feasible and practical policy balanced, and reflective. It is a student-centric programme which consists of 3
changes by which heuristics could be used to benefit student learning and
core concepts and 6 groups of subjects which are taken by every IBDP Student.
performance.
The three core concepts are CAS (Creativity, Activity, Service) which ensures
that students are exposing themselves to activities other than academics, TOK
1 Introduction (Theory of Knowledge) which allows students to question what knowledge is,
where it comes from and how its applied, and the EE (extended essay) which
Behavioural economics is a branch of economics which considers psychology and is a 4000 word research report written by each student in a subject of their
recognises that choices made by consumers are based on various factors that may preference and on a topic of their choice. The IB has six groups of subjects as
or may not agree with the classical or neoclassical approach to economics. The seen in figure 1 below5 .
field studies and describes economic decision-making. According to theories of 2 “Heuristic.” BehavioralEconomics.com — The BE Hub, 28 Mar. 2019,
behavioural economics, actual human behaviour is less rational, stable, and www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/heuristic.
3 Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioural eco-
selfish than traditional normative theory suggests, due to limited rationality and
self-control, as well as social preferences.1 nomics. The American Economic Review, 93, 1449-1475
4 Iborganization. “Diploma Programme (DP).” International Baccalaureate®,
∗Advised by: Dr Edoardo Gallo, Assistant Professor and Director of Studies (Economics) www.ibo.org/programmes/diploma-programme/.
5 “IB Subject Briefs: St. Andrews International School Bangkok.” IB Subject Briefs
at University of Cambridge
— St. Andrews International School Bangkok, www.nordangliaeducation.com/our-
1“Behavioral Economics.” BehavioralEconomics.com — The BE Hub, 23 Oct. 2018, schools/bangkok/learning/curriculum-overview/high-school/international-baccalaureate/ib-
www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/behavioral-economics. diploma-subject-briefs.
30
29
3 Status Quo Bias
“When faced with a choice among different options, people have a tendency to
stick with the default”. 6 This is known as the status quo bias. It is caused by
loss aversion or the belief that “losses loom larger than gains” 7 and is affected by
choice – the bias is stronger when more choices are available or when the choices
available are more complex. To clarify, the status quo bias 8 is witnessed when
people choose to make the same decision they had made previously without
logically and fairly considering other choices or when individuals resist change.
9
The status quo bias impacts several aspects of education. One significant
impact of this bias is on the methods by which teachers teach their classes. Over
the past decade, several research studies10 have proven that students learn best
in different ways and prefer certain teaching resources that are based on their
preferred style of learning. However, it is often seen that teachers teach their
classes in the style that is preferable to them rather than their students11 . This
is the status quo. Because of the status quo, a hypothetical teacher who prefers
visual learning may use charts and diagrams to explain key terms and processes.
This would be ideal for visual learners who grasp concepts easier and faster when
displayed visually; however, it would not be optimal for auditory learners (learn-
Figure 1: IB Subject Groups
ers who prefer learning through sound) or for kinaesthetic learners (learners who
prefer to learn through movement and touch). So, if this hypothetical teacher’s
Group 1 - Studies in Language and Literature (subjects include language class were comprised of 90 percent of students being auditory or kinaesthetic
and literature, literature) learners, the visual teaching style would not be beneficial. To generalise, if
Group 2 - language acquisition - learning a second or foreign language (sub- the teaching style employed by a teacher closely matches a student’s preferred
jects include hindi, spanish, french, italian, etc.) style of learning, learning becomes easier, faster and relatively effortless for the
Group 3 - individuals and societies - social sciences (subjects include history, student, and if it does not, learning becomes more time consuming, relatively
geography, psychology, economics, etc.) difficult, and rather arduous.12 If the teacher’s style is beneficial for a large
Group 4 - Sciences (subjects include physics, chemistry, biology, etc.) majority of students in their class it is the best possible, or ideal, teaching style
Group 5 - Mathematics (analysis and approaches or analysis and interpre- and the status quo is working effectively; however, if it does not agree with the
tation) learning style of many students, it is ineffective and needs change.
Group 6 - Arts (subjects include art and design, music, theatre, etc.) Teaching in a style which is beneficial to a majority of students is encouraged
Most IBDP students study three subjects at the higher level and three at 6 Thaler, Richard H., and Cass R. Sunstein. Nudge: Improving Decisions about Health,
the standard level. Each student must take one subject from groups 1-5 (each) Wealth, and Happiness. Yale University Press, 2008.
7 Kahneman, D., and Tversky, A. (1982). The psychology of preference. Scientific Ameri-
and as their 6th subject they may choose to take a form of the arts or a second
can, 246, 160-173.
subject from either group 3 or group 4. Each student is given a final score out 8 “Status Quo Bias.” BehavioralEconomics.com — The BE Hub, 28 Mar. 2019,
of 45 at the end of this two-year programme. Each student is awarded a certain www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/status-quo-bias/.
number of points from 1 to 7 (7 being the highest which is normally equivalent 9 Samuelson, W., and Zeckhauser, R. J. (1988). Status quo bias in decision making. Journal
to an A or A* in IGCSE) for each of their 6 subjects. This compromises 42 of Risk and Uncertainty, 1, 7-59.
10 Franzoni, Ana Lidia, et al. “Student Learning Styles Adaptation Method Based on
of the 45 available IB points. The three remaining points are taken from the
Teaching Strategies and Electronic Media.” 2008 Eighth IEEE International Conference on
extended essay and theory of knowledge assessments. CAS is not evaluated but Advanced Learning Technologies, 2008, doi:10.1109/icalt.2008.149. See section on “related
is necessary for the successful completion of the IBDP. works”.
11 Franzoni, Ana Lidia, et al. “Student Learning Styles Adaptation Method Based on Teach-
ing Strategies and Electronic Media.” 2008 Eighth IEEE International Conference on Ad-
vanced Learning Technologies, 2008, doi:10.1109/icalt.2008.149.
12 Rose, C. (1998). Accelerated Learning, New York: Bantam Dell Publishing Group.
31 32
hundreds of researchers who agree that teaching styles of teachers should be
based on their students’ preferred learning styles, as opposed to being a reflection
of the teacher’s ideal working style, but several problems – including learning
ability, background knowledge, learning goals, and learning style of students
– need to be overcome in order to reach a state where teaching styles can be
easily optimised for their audience13 In an effort to solve these problems and
propose a method by which teaching can be modified for student benefit, Ana
Lidia Franzoni and Saı̈d Assar, the co-authors of the study ‘Student Learning
Styles Adaptation Method Based on Teaching Strategies and Electronic Media14
examined the learning styles model by Felder-Silverman.
The Felder-Silverman learning styles model15 examines students’ learning
styles based on four parameters:
1. What kind of information does the student prefer to receive?
a. sensitive – prefers concrete, practical, and fact-based thinking and ap-
proaches Figure 2: Adaptive Teaching Taxonomy relation entity diagram
b. intuitive – prefers conceptual thinking and approaches such as theories
2. Through which sensorial channel does the student receive information
more effectively? well. Their summarised learnings of recommended teaching styles and elec-
a. visual representations (diagrams and charts) tronic methods can be seen in figures 3 and 4, respectively.16
b. verbal (spoken words, written explanations)
3. How does the student process information?
a. actively (groupwork, discussions, activities) or
b. reflexively (through introspection and reflection)
4. How does the student make progress?
a. sequentially (learns in small steps of increasing complexity), or
b. globally (learns through holistic thinking and prefers learning all, or most,
of a concept at once)
These four parameters were analysed by Franzoni and Assar who used their
results to create a metric using descriptions, ideal pedagogical methods, and
characteristics of media to explain the ideal teaching styles and use of electronic
methods based on students’ learning styles (see figure 3).
For example, they found that visual learners relied heavily on visual com-
ponents and tended to do better on assignments that asked them to visualise Figure 3: Adaptive Taxonomy
or explain concepts using visual representations. Specifically, they found that
images, diagrams, and charts helped students understand and summarise their While this study provides a logical sounding proposal to mitigate the status
learnings. Based on these observations, Assar and Franzoni proposed that vi- quo bias by matching teaching and learning styles, where they do not already,
sual learners be taught using strategies and games, and that electronic media it is important to understand that this study only proposes a potential model
like electronic presentations, videos, and animations be used as learning aids. and further research through experimental studies is needed to ensure that the
The two researchers repeated this process for several other learning types as model does, in fact, work and to prove a definite causal relationship between
teaching styles and improvements in student performance. It is also important to
13 Ford, N., and Chen, S. (2001). Matching/mismatching revisited: an empirical study of
mention that there are several counter-arguments to teaching based on learning
learning and teaching styles. British Journal of Educational Technology, 32 (1), 5-22.
14 Franzoni, Ana Lidia, et al. “Student Learning Styles Adaptation Method Based on Teach- styles as contradictory studies17 have found no substantial benefit of modifying
ing Strategies and Electronic Media.” 2008 Eighth IEEE International Conference on Ad- 16 Franzoni, Ana Lidia, et al. “Student Learning Styles Adaptation Method Based on Teach-
vanced Learning Technologies, 2008, doi:10.1109/icalt.2008.149. ing Strategies and Electronic Media.” 2008 Eighth IEEE International Conference on Ad-
15 “Felder-Silverman.” The Peak Performance Center, www.thepeakperformancecenter.com/educational-
vanced Learning Technologies, 2008, doi:10.1109/icalt.2008.149.
learning/learning/preferences/learning-styles/felder-silverman/. 17 Kevin Donnelly Senior Research Fellow - School of Education. “’Chalk and
33 34
they would be able to stand or move around while learning. The mental status
quo of sitting while learning could also be removed through having standing
desks. Students could be allowed to take notes both digitally and by hand – as
they prefer. This would allow visual learners to add images and diagrams to
their notes with ease. Lastly, different teachers teach in different ways so if there
are multiple batches of students in a grade who are studying the same subject,
recordings of the classes could be shared with all students taking a given sub-
ject. This would be beneficial not only to auditory learners, who can play-back
the recordings, but also to all learners to get access to teaching styles that may
be closer to their preferred learning style. It is important to note that there
are certain ethical considerations that must be thought of with regards to the
recording of classes, before they are shared with students. A waiver for privacy
and the ethical use of recordings could be sent to parents or legal guardians,
of students, to ensure responsible data use. Allowing for these small changes
in learning space and accessible resources would make it far easier for students
to learn as they wish to and for teachers to teach in their preferred style while
simultaneously reducing the status quo bias that exists between teaching and
learning styles.
As stated by Liz Bergeron and Michael Dean in their study, “The IB Teacher
Professional: identifying, measuring and characterising pedagogical attributes,
perspectives and beliefs”, one of the key features of any successful IBDP teacher
is flexibility or adaptability 19 . This is because the IBDP reviews all of its pro-
gram syllabi every seven years in an attempt to “ensure that each (subject) is fit
for purpose in a changing world” and that each subject’s syllabus “incorporates
the latest educational research and lessons learned from a thorough evaluation
Figure 4: Adaptive Taxonomy of the existing curriculum. 20 This frequent change in syllabus is often the
second main cause for the status quo bias impacting education.
teaching styles to match student’s preferred learning styles18 Overall, this study The prevalence of the status quo bias due to syllabus changes can be com-
by Assar and Franzoni is a great example of how the status quo surrounding pared to an example of a healthcare study done by Samuelson William and
learning and teaching styles could be mitigated where needed. The study is also Richard Zeckhauser in their research paper “Status Quo Bias and Decision
a good example for teaching styles IBDP teachers could use as it relates well to Making”21 . In the study, which was based on field data from Harvard Uni-
the holistic approach to education and the learning objectives that the IBDP versity’s health plan enrolments of 9,185 employees in 1986, a large disparity
aims for its students to achieve. was observed between the health plan choices of new and old enrolees. Each
To understand their students’ learning styles and adapt their teaching styles year, enrolees were allowed to transfer from one health plan to another at no
accordingly, teachers could use the Felder-Silverman learning styles model, posed transaction cost. It was found that enrolees who joined the program in 1986
in the form of a questionnaire. Adapting learning experience to learning styles in preferred to stick with the plan they had originally chosen as compared to a new
order to reduce the status quo bias can be done through several other methods as plan which had more favourable premiums and deductibles. This trend was not
well. Classrooms could have various seating or non-seating options for students observed with new enrolees who joined the program in 1987, thus displaying the
to stay comfortable while they learn. For this, classrooms could have regular prevalence of the status quo bias in old enrolees – despite having no transaction
desks, standing desks, floor-space, bean bags, etc. so that students can choose cost (normally the main reason participants remain with the status quo is due
how they learn. This would be especially beneficial for kinaesthetic learners as 19“IBO Publication.” Ibo.org - Global Assets,www.ibo.org/globalassets/publications/ib-research/
continuum/theibteacherprofessional_final_march6.pdf
Talk’ Teaching Might Be the Best Way after All.” The Conversation, 12 May 2020, 20Iborganization. “Latest Curriculum Updates.” International Baccalaureate®,
www.theconversation.com/chalk-and-talk-teaching-might-be-the-best-way-after-all-34478.
18 Rogers, Vincent, and Joan Baron. Â Teaching Styles and Pupil Progress.â The Phi www.ibo.org/university-admission/recognition-of-the-ib-diploma-by-countries-and-
Delta Kappan, vol. 58, no. 8, 1977, pp. 622â 623. JSTOR, www.jstor.org/stable/20298722. universities/latest-curriculum-updates/.
Accessed 30 Aug. 2020 21 Samuelson, W., and Zeckhauser, R. J. (1988). Status quo bias in decision making. Journal
of Risk and Uncertainty, 1, 7-59.
35
36
to transaction costs as human beings practise loss aversion). Evidence of these work being scored less as the teacher’s criteria – which is not a necessity by the
findings can be seen in figure 5 (below). IBDP – is not followed. To improve on this, the IBDP should have free, virtual
teacher training workshops for teachers affiliated with IBDP schools explaining
syllabus changes and how they can be adapted to with ease, as well as clarifying
any doubts that teachers may have. The IBDP should also have a helpline for
teachers to call so that they can clarify any doubts relating to syllabi, clearly
and directly.
4 Conformity Bias
As defined by the McCombs School of Business, conformity bias is the “ten-
dency of individuals to behave like those around them instead of using their
own judgement”23 . It relates to the behavioural aspect of herding, which has
been a well-known concept in both psychology and philosophy for a centuries.
Herding behaviour24 is when people follow what those around them are doing
Figure 5: Healthcare Plan Transfers instead of making their own individual and independent decisions, based on
the information which is available to them. Herding behaviour is influenced by
Similar to what is seen in the study above, is the effect of the status quo several factors: fear, uncertainty, the shared identity of decision makers, etc.
bias of teachers with regards to syllabus changes. Due to extremely frequent This behaviour, of conformity bias and herding, could be both beneficial and
syllabus changes in the IBDP, teachers often have to simultaneously teach dif- dangerous depending on the context in which it is used. As such, it could be
ferent syllabi of the same subject to two year groups. This may lead to teachers extremely beneficial for the education sector, but only if used effectively.
being confused and struggling to remain clear on the syllabi of each year group. To test whether conformity bias is seen in the extracurricular activities of-
In turn, may also lead to teachers preferring previous syllabi and books as they fered by schools or undertaken by students, a simple google search is adequate.
understand those better and have greater understanding of their subject mat- On the 27th of August, 2020 at 6:43pm I googled the following phrases, “popular
ter. In this case they may continue teaching past syllabi, which may affect the extracurriculars in schools” and “the best extracurricular activities for college”
performance of students (who have an adapted syllabus). Moreover, the status and opened up the most popular 8 sites that appeared (the 8 sites that appeared
quo bias may be responsible for teachers choosing not to adapt their courses or closest to the top of the search with both key phrases). Below are the websites.
teaching techniques to meet the requirements of the new syllabi. Essentially, Note: the reliability of the websites was not checked, they were simply the
teachers may not adapt their teaching style to fit the new syllabus thereby ones that appeared closest to the search bar with both sets of key terms. The
affecting students’ performance, especially as the question types in exam pa- findings drawn from these websites (described below) are not hard evidence and
pers often change with syllabus changes. It is important to note that further their results need to be replicated through the use of experimentation and data
quantitative data analysis is required to validate these logical correlations and gathering. They were the 8 most frequently visited websites that matched both
ensure the transferability between healthcare data and adaptations of teachers selected key phrases.
to changes in syllabi. The results yielded from these websites were almost identical to each other.
It is also important to consider cultural contexts with regards to the status All the websites focused on displaying ‘leadership, passion and impact’ or an-
quo bias. In several countries like India and China, local curricula are tend other variation of those same three words. To achieve these three goals, they
towards being more structured and less subjective than the IBDP22 . Because of recommended certain activities as well: academic clubs, the arts and sports were
this, teachers teaching the IBDP in these nations may get intimidated by the spoken about generally and Model United Nations (MUN), student government,
curriculum’s relative lack of structure. This may result in teachers creating a debate club, yearbook, robotics, internships and volunteer work, specifically,
specific format based on previous answers that did well – a direct effect of the were some common extracurriculars listed. Some websites did provide extensive
status quo bias. This is especially common for internal assessments and the lists of popular extracurriculars with a greater number of suggestions, however,
extended essay which are in essay format. These preconceived notions of model 23 “Conformity Bias.” Ethics Unwrapped, 12 Dec. 2018,
answers, held by teachers, may lead to children with otherwise well-written ethicsunwrapped.utexas.edu/glossary/conformity-bias.
24 “Herd Behavior.” BehavioralEconomics.com — The BE Hub, 29 Mar. 2019,
22 The Swaddle. “How Learning Differs Across CBSE, ICSE, IB and CIE Education Boards.”
www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/herd-behavior/.
The Swaddle, 15 May 2018, theswaddle.com/learning-education-boards/.
37 38
these were some of the most frequently observed results. The fact that some the importance of doing what interests a student through voicing the negative
results were found more frequently than others suggests that these activities are impacts a lack of individuality could have on an individual. Additionally, schools
extremely popular with students and may even be prioritised by schools. Thus should create lists of locally available extracurricular activities with details so
this is clear evidence of the conformity bias at work. that students can easily explore a range of interesting activities. This database
The prioritising of certain extracurriculars can have a vast range of negative could be based on the extracurriculars done by previous students in the school.
impacts. Prioritised extracurriculars become extremely popular in schools and Lastly, a new activity could be introduced in the school each year or each month
may be given more importance and funding than other activities. While this is to give students exposure to new creative outlets and skills that could help them
beneficial to some students who truly enjoy and are particularly skilled at these through life. Examples include a cookery club or a financial literacy workshop.
activities, it may hamper the holistic growth of other students who have different The lack of conformation bias may also be problematic in educational in-
interests. As more and more students use the conformity bias and herd to stitutions. When different teachers teach batches of a grade the same subject,
popular activities, less priority is given to other extracurriculars and soon people some groups of students may learn more than others due to the different teach-
stop opting for and exploring them, leading to their removal from the list of ing styles of teachers27 . This lack of conformation occurs as the IBDP syllabus is
activities offered. This limits extracurricular choices for students in later years. subjective (comparative to other curricula) and encourages various approaches
Moreover, because these activities are prioritised, students feel peer pressure to to both teaching and learning28 . As the syllabus has a broad scope, teachers
participate in them for fear of missing out. Another important consequence of may teach different things. This lack of conformity may negatively impact the
prioritising certain extracurriculars is that the conformity bias leads to them results of some students while unfairly advantaging others due to their teacher’s
no longer being unique or special to students when applying to college. So, teaching style.
extracurriculars which would otherwise make students competitive and stand- Consider a realistic hypothetical situation wherein there are three batches
out on college applications no longer do. This puts pressure on students to of an Economics Higher Level class for 11th grade students at School X in a
not only follow others and take part in the prioritised extracurriculars to be given year (say 2020), and each of these three classes are taught by different
competitive, but also pushes them into taking part in additional extracurriculars teachers. Assume that teacher A who teaches Batch 1 teaches through real
to make them stand out. As Columbia University psychology professor, Dr. life examples but does not teach students key economic concepts with formal
Suniya Luthar says, children get extremely caught up in extracurriculars and technology, Teacher B who teaches batch 2 explains economic concepts and
try to do everything often forgetting where to stop. This becomes an issue defines key terms but does not cover real life examples in depth, and Teacher
when the child starts to say “his or her performance determines his or her self- C who teaches batch 3 does both. This information is summarised in Figure
worth: I am as I perform”. She believes that an increasing number of children 7. If all three batches are given the same exam, in which they are asked to
are associating self-worth with their academic performance and involvement explain a key concept using definitions, key terms and relevant, detailed real life
in extracurricular activities and that this is extremely dangerous25 . Due to examples, Batch 3 would be unfairly advantaged as Batch 1 would not be able
additional pressures to stand out, students’ mental and physical health may be to give a detailed real life example and Batch 2 would not be able to define or
harmed as they may battle with anxiety, obesity from lack of physical exercise name key terms while Batch 3 would be able to do both. This would, thus, be
(caused by a lack of time), chronic stress at a young age, feeling lost, depression, a negative consequence of the lack of conformity between teachers and would
feeling overwhelmed and getting panic attacks, etc. Essentially, though schools negatively impact the performance of students.
may prioritise certain extracurriculars to make their students competitive, this In an attempt to reduce the negative impacts of the lack of conformity,
prioritisation may have several negative consequences on students. recordings could be taken of each of the batches’ sessions and be made available
In the case of extracurriculars, confirmation bias – when people try to find to all students taking that subject. In addition, teachers could collaborate and
and analyse data such that it fits with their existing preconceptions26 – along make a combined lesson plan for the year to ensure the same content, at least
with the fear and uncertainty that is part of the college application process may in terms of concepts, is covered.
be the cause for conformation bias. This is because students tend to assume that Another area of education that may be affected by the conformation bias
the only way to get to a certain institution is by doing exactly what someone, is somewhat contrasting to the previous point on the disadvantages of teachers
who got accepted into that institution previously, did. having different teaching methods. It is when teachers imitate or copy other
To reduce conformation bias in extracurriculars, teachers should explain teachers’ teaching styles. This imitation may be due to various factors such as
27 Franzoni, Ana Lidia, et al. “Student Learning Styles Adaptation Method Based on Teach-
25 Ruthie, AAE. Association of American Educators,
www.aaeteachers.org/index.php/blog/1154-how-far-is-too-far-students-and-extra-curricular- ing Strategies and Electronic Media.” 2008 Eighth IEEE International Conference on Ad-
activities. vanced Learning Technologies, 2008, doi:10.1109/icalt.2008.149.
26 “Confirmation Bias.” BehavioralEconomics.com — The BE Hub, 29 Mar. 2019, 28 “International Baccalaureate (IB) Program / Approaches to Learning (ATL).” / Ap-
www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/confirmation-bias/.
proaches to Learning (ATL), www.bostonpublicschools.org/Page/5935.
39
40
seniority and experience of a certain teacher, or s tudents of a certain teacher’s compared to a hundred and ten dollars tomorrow, but would not mind waiting
batch performing consistently better than other batches. W hile copycat bean extra day if the choice were for the same amounts one year from today versus
haviour29 by teachers may be beneficial in terms of ensuring that all s tudents get one year and one day from today. In a general sense, the concept is often used to
the s ame learning in terms of content and teaching method, it has s everal describe impatience or the preference for immediate gratification in an individ-
disadvantages. Copycat behaviour or imitation may lead to teachers being out of ual’s decision-making34 . In ‘Present-Bias, Procrastination and Deadlines in a
their comfort zone and therefore not being able to impart education in the most Field Experiment’35 , a study done by Alberto Bisin and Kyle Hyndman, it was
successful manner – leading to s tudents having difficulties learning and found that subjects who face an absolute or set deadline start working on tasks
comprehending new concepts. Moreover, the lack of variety in teaching s tyles, earlier than subjects who do not have a definite or imposed deadline (Panel
caused by imitation, would logically increase the chance that a teacher’s method A of Figure 8). Yet, contrary to expectations, it was also noted that subjects
of teaching does not r elate to a s tudents preferred learning s tyle. Consider-ing the with a definite deadline began working on tasks far closer to their deadline as
evidence s upporting better comprehension and performance of s tudents when compared to subjects who did not have a binding, approaching deadline (Panel
teaching is catered to their preferred learning method30 ( discussed under the B on Figure 8). Lastly, it was observed that subjects who began working on a
status quo effect) this may also be a major disadvantage of teaching s tyle task closer to its deadline were less likely to complete it than those who began
imitation. earlier (Panel C on Figure 8). In fact, students who began their task on the same
A potential s olution to imitation could be the encouragement of various day as its deadline has a less than 50 percent chance of completing the task,
teaching s tyles31, as endorsed by the I BO ( International Baccalaureate Organ- and those who began a week, or more, before their deadline had a 70 percent
isation), paired with s hared audio r ecordings of classes s o that s tudents can or higher probability of completing it. Arguably, the most interesting finding
benefit f rom teaching s tyles of various teachers. ( This policy change has been of the study was that each additional day before the deadline that a task was
explored f urther in the previous s ection of the r eport – the s tatus quo effect). started, the probability of it being completed increased by 2.6 percent. As such,
While the discussions, on both copycat behaviour and on variety in teaching it could be concluded that binding deadlines help ensure tasks are completed.
styles, are based on logical assumptions which explore a valid area in which While this study in itself proves the existence of the present bias and its
education in the IBDP may be negatively impacted by the conformation bias, impact on education, it is worth noting that the results of this study prove
both methods require further research in terms of quantitative evaluation lend the idea that humans underestimate the amount of time they need to complete
them credibility and reliability. disjunctive – independent – tasks as they are increasingly unable to map time
accurately the further away the events are from present time. There are several
negative consequences of mankind’s inability to map future time accurately, on
5 Present Bias education.
As teachers are human, they tend to slow down their teaching speed in
In psychological literature, procrastination is normally defined as the practice of
the beginning of an academic year while revising the basic topics or previously
pushing impending tasks to a later time or date when s uch a task r esults in a
studied material and then later face a lack of time and rush through some
“counterproductive and needless delay”. I t is generally considered to be a r esult
of the most complex parts of the syllabus in order to meet their set syllabus
of present-bias in psychology and economics, and r esults in an individual delaying
completion date. This effect of the present bias may make it difficult for students
unpleasant tasks that, in hindsight, they wish they would have com-pleted
to understand tougher, newer concepts, which are normally taught towards the
sooner32. ‘ The present bias r efers to the tendency of people to give more end of an academic year. Moreover, when paired with the status quo effect, the
consideration and weight to events and r ewards that are closer to the present
present bias also leads to teachers spending more time on the syllabus content
time when considering differences between two f uture moments’33. For exam-ple, that is simplest for them to teach as they can explain it with clarity. However,
a present-biased person might prefer to r eceive a hundred dollars today as
this also means that teachers brush through complex concepts as a consequence
29“COPYCAT: Meaning in the Cambridge English Dictionary.” Cambridge Dictionary, dic- – once again impacting students’ understanding of complex concepts. In the
tionary.cambridge.org/dictionary/english/copycat. case that teachers do not complete syllabus on time, students may get less time
30Franzoni, Ana Lidia, et al. “Student Learning Styles Adaptation Method Based on Teach- to study for their exams. This would impact their progress, ability to revise
ing Strategies and Electronic Media.” 2008 Eighth IEEE International Conference on Ad- concepts in a timely manner, and lead to them not having enough time to
vanced Learning Technologies, 2008, doi:10.1109/icalt.2008.149. prepare well for their tests or exams – possibly affecting their performance. On
31“International Baccalaureate (IB) Program / Approaches to Learning (ATL).” / Ap- 34 “Present Bias.” BehavioralEconomics.com — The BE Hub, 28 Mar. 2019,
proaches to Learning (ATL), www.bostonpublicschools.org/Page/5935. www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/present-bias/.
32Bisin, Alberto, and Kyle Hyndman. “Present-Bias, Procrastination and Deadlines in a 35 Bisin, Alberto, and Kyle Hyndman. “Present-Bias, Procrastination and Deadlines in a
Field Experiment.” 2014, doi:10.3386/w19874. Field Experiment.” 2014, doi:10.3386/w19874.
33O’Donoghue, T., and Rabin, M. (1999). Doing it now or later. American Economic
Review, 89(1), 103-124. 42
41
May 3rd, 2000, the BBC News published an article titled “Unfinished courses bring students of his economics class at Massachusetts Institute of Technology (MIT)
exam panic” which contained quotes by s tudents f rom across England on the lack as his subjects. Each student had to complete 3 assignments within 12 weeks,
of s yllabus completion and its impact on their ability to s tudy f or their board however, the deadlines for each batch of students differed. Batch 1 was told
examinations. One s uch quote r ead “Our English teacher hasn’t finished our that they had no deadline; they just had to submit all three papers by the end
syllabus yet!!!!! That’s why I ’m here trying 2 do it myself!!!! How do they expect of the semester. Batch 2 was told that they could choose any deadline for each
you to pass ya exams when they don’t even teach you!!!” while another read, paper (with the last possible deadline being the last day of the semester), but
“with now 12 dayz to go till my I T exam, s uicide is becomin a very attractive that they had to submit their chosen deadlines to Ariely by within 5 days of
alternative and no teenz s hould f eel like we do ...”36. As s uggested by the the start of the semester and would face penalties of increasing value each day
anecdotal evidence, non-completion of s yllabi, caused by the present bias, may that their paper was late (from their chosen deadline). Batch 3 was given fixed
lead to s evere consequences including excessive s tress, confusion, and mental deadlines: Paper 1 was due at the end of Week 4, Paper 2 at the end of Week
health disorders in s tudents. 8, and Paper 3 at the end of Week 12. There would be no benefit of submitting
While considering the s ubstantial negative impacts, of teachers s lowing their work early for any group. When Ariely assessed the students, oblivious to who
pace initially and increasing it towards the end of the academic year, it is worth was in which batch, then analysed the results, he found that Batch 3 had done
noting the Blooms Taxonomy37 ( figure 9), the f oundation of many educational the best, followed by Batch 2, and Batch 1 had done the worst. Thus, the study
courses and teachers’ lesson plans, explains one r eason why this is done. Accord- concluded that external influence was far more powerful at ensuring deadlines
ing to the taxonomy, teachers s pend time on initial chapters to help s tudents were met as compared to self-control and self-motivation. The practise of setting
recall what they have previously learnt and to explain the methods in which the external deadlines could be an effective method for ensuring timely syllabus
course would progress. Towards the end of the year, s tudents are f amiliar with the completion and lowering of the present bias. This may be achieved in several
methods in which to approach concepts and questions, thus teach-ers can speed ways, for example, teachers could create a detailed checklist for everything that
through the s yllabus and f ocus more on the aspects of analysis, evaluation, and is to be done. This checklist could be shared with students as chapters are
creativity. Essentially, the taxonomy s tates that teachers move through their completed (to prevent choice overload by sharing the entire syllabus at once)
syllabi s lowly, at first, not due to the present bias, but r ather to help s tudents to ensure that all essential aspects of the syllabus are completed. This would
learn and comprehend f oundational information. W hile the existence of this help both students and teachers understand the extent of syllabus that has been
taxonomy r educes the credibility of the link between present bias and teaching covered. Furthermore, adding estimated dates for when a certain part of the
speed, it is a valid counterargument and thus must be consid-ered. Furthermore, syllabus should be completed would help ensure that the syllabus is completed
the existence of the taxonomy does not mitigate the impact that present bias does in a timely manner. This is because it would transform the syllabus from a
have on teaching s peed. Despite creating lesson plans with the help of the single two-year long disjunctive event to a series of conjunctive events (a series
taxonomy, teachers do tend to r un s hort of time and s ome-times s truggle to of events that come one after another). This would be beneficial as humans
complete s yllabi at the end of an academic year – leading to a potential decrease tend to overestimate the amount of time they need to complete a series of small
in s tudent performance. events.
Present bias also affects s tudents. W hen s tudents r eceive their exam time-
tables thirty days prior to their first exam, they are not able to accurately
visualise the time they have to r evise their s yllabus. This s ame trend may be noted 6 Conclusion
when I BDP s tudents have deadlines f or internal assessments, lab r eports, and
Education leads to the creation of an intelligent and competent future gener-
their extended essay due months after they are notified ( of the deadlines). As such,
ation. As such, imparting the best possible, or optimal, education, is key to
procrastination on these assignments, due to the present bias, would also impact
delivering an intelligent and competent generation. The IBDP (International
the r esults of s tudents.
Baccalaureate Diploma Programme), which strives for holistic learning and aims
An experiment conducted by Dan Ariely, mentioned in Chapter 6 of his book
to produce young adults who are aware of global affairs, sets out on this mission.
‘Predictably I rrational’38, explains potential method by which the effect of the
However, heuristics, or systemic biases, as defined by the field of behavioural
present bias could be lowered. I n this experiment, Ariely used three batches of
economics, may impact education negatively. Heuristics that tend to have a
36“EDUCATION —Unfinished Courses Bring Exam Panic.” BBC News, BBC, 3 May 2000, great impact on education include the status quo effect, the conformation bias,
news.bbc.co.uk/2/hi/uk news/education/733362.stm. and the present bias. Findings related to each of these heuristics have been
37Mcdaniel, Rhett. “Bloom’s Taxonomy.” Vanderbilt University, Vanderbilt University, 25 discussed in detail within the research paper.
This paper is based solely on observation and secondary data. Some of the
Mar. 2020, cft.vanderbilt.edu/guides-sub-pages/blooms-taxonomy/.
studies mentioned in the paper need further research and evaluation to improve
38Ariely, Dan. Predictably Irrational: the Hidden Forces That Shape Our Decisions. Harper
Perennial, 2010.
44
43
7.2 Websites
“Behavioral Economics.” BehavioralEconomics.com — The BE Hub, 23 Oct.
2018,
the validity of their results. Additionally, the proposals in this study need to www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/behavioral-
be tested to examine whether or not the suggestions have a positive impact on economics.
student learning and performance. That being said, this paper considers sev- “Confirmation Bias.” BehavioralEconomics.com — The BE Hub, 29 Mar.
eral sources of information and their application and is written with first-hand 2019, www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/confirmation-
knowledge of the IBDP. It presents logical arguments and supports them with bias/.
previously performed studies. It also considers counterarguments and exam- “Conformity Bias.” Ethics Unwrapped, 12 Dec. 2018, ethicsunwrapped.utexas.edu/
ples where necessary. Lastly, it proposes a wide-range of methods by which glossary/conformity-bias.
the learning and performance of IBDP students may be improved. Thus, while “COPYCAT: Meaning in the Cambridge English Dictionary.” Cambridge
further research is recommended in the subject area, it is reasonable to state Dictionary, dictionary.cambridge.org/dictionary/english/copycat.
that the paper provides feasible and logical methods by which the performance “EDUCATION — Unfinished Courses Bring Exam Panic.” BBC News, BBC, 3 May
of IB students may be improved. 2000, news.bbc.co.uk/2/hi/uk news/education/733362.stm.
The paper concludes that status quo bias, conformity bias, and present bias “Felder-Silverman.” The Peak Performance Center,
do affect the formal education system and that, when employed correctly, their www.thepeakperformancecenter.com/educational-learning/learning/preferences/
influence can be used to positively impact the learning and performance of IBDP learning-styles/felder-silverman/.
students.
“Herd Behavior.” BehavioralEconomics.com — The BE Hub, 29 Mar. 2019,
www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/herd-behavior/.
7 Bibliography “Heuristic.” BehavioralEconomics.com — The BE Hub, 28 Mar. 2019,
www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/heuristic.
7.1 Books and Journals Iborganization. “Diploma Programme (DP).” International Baccalaureate®,
www.ibo.org/programmes/diploma-programme/.
Ariely, Dan. Predictably Irrational: the Hidden Forces That Shape Our Deci-
Iborganization. “Latest Curriculum Updates.” International Baccalaure-
sions. Harper Perennial, 2010.
ate®, www.ibo.org/university-admission/recognition-of-the-ib-diploma-by-countries-
Bisin, Alberto, and Kyle Hyndman. “Present-Bias, Procrastination and
and-universities/latest-curriculum-updates/.
Deadlines in a Field Experiment.” 2014, doi:10.3386/w19874.
“IBO Publication.” Ibo.org - Global Assets, www.ibo.org/globalassets/publications/ib-
Franzoni, Ana Lidia, et al. “Student Learning Styles Adaptation Method
research/continuum/theibteacherprofessional_final_march6.pdf
Based on Teaching Strategies and Electronic Media.” 2008 Eighth IEEE Interna-
tional Conference on Advanced Learning Technologies, 2008, doi:10.1109/ “IB Subject Briefs: St. Andrews International School Bangkok.” IB Subject
icalt.2008.149. Briefs — St. Andrews International School Bangkok, www.nordangliaeducation.com/
our-
Ford, N., and Chen, S. (2001). Matching/mismatching revisited: an em- schools/bangkok/learning/curriculum-overview/high-school/international-
pirical study of learning and teaching styles. British Journal of Educational baccalaureate/ib-diploma-subject-briefs.
Technology, 32 (1), 5-22. “International Baccalaureate (IB) Program / Approaches to Learning (ATL).”
Kahneman, D. (2003). Maps of bounded rationality: Psychology for be- / Approaches to Learning (ATL), www.bostonpublicschools.org/Page/5935.
havioural economics. The American Economic Review, 93, 1449-1475. Kevin Donnelly Senior Research Fellow - School of Education. “’Chalk
Kahneman, D., and Tversky, A. (1982). The psychology of preference. Sci- and Talk’ Teaching Might Be the Best Way after All.” The Conversation,
entific American, 246, 160-173. 12 May 2020, www.theconversation.com/chalk-and-talk-teaching-might-be-the-
best-way-after-all-34478.
O’Donoghue, T., and Rabin, M. (1999). Doing it now or later. American
Economic Review, 89(1), 103-124. Mcdaniel, Rhett. “Bloom’s Taxonomy.” Vanderbilt University, Vanderbilt
Rose, C. (1998). Accelerated Learning, New York: Bantam Dell Publishing University, 25 Mar. 2020, cft.vanderbilt.edu/guides-sub-pages/blooms-taxonomy/.
Group. “Present Bias.” BehavioralEconomics.com — The BE Hub, 28 Mar. 2019,
Samuelson, W., and Zeckhauser, R. J. (1988). Status quo bias in decision www.behavioraleconomics.com/resources/mini-encyclopedia-of-be/present-bias/.
making. Journal of Risk and Uncertainty, 1, 7-59. Rogers, Vincent, and Joan Baron. Â Teaching Styles and Pupil Progress.â
Thaler, Richard H., and Cass R. Sunstein. Nudge: Improving Decisions The Phi Delta Kappan, vol. 58, no. 8, 1977, pp. 622â 623. JSTOR, www.jstor.org/
about Health, Wealth, and Happiness. Yale University Press, 2008. stable/20298722.
Ruthie, AAE. Association of American Educators, www.aaeteachers.org/
index.php/blog/1154-how-far-is-too-far-students-and-extra-curricular-activities.
“Status Quo Bias.” BehavioralEconomics.com — The BE Hub, 28 Mar.
46
45
Figure 7: A Realistic Hypothetical Situation of Non-Conformity
Figure 6: Extracurricular Advice Websites
Figure 8: The time of task insurance
47 48
Oncogenic Viruses: A Potential CRISPR
Treatment
∗
Neha Matai
April 2, 2021
Abstract
Oncogenic viruses promote carcinoma development by establishing
long-term latent infections, obstructing tumor suppressor pathways, and
transforming host cells into unchecked, proliferating malignancies. Epstein-
Barr Virus, the first oncogenic virus to be discovered, can promote lym-
phomagenesis in T cells, NK cells, and most commonly, resting memory
B cells through the expression of oncoprotein LMP-1 and other proteins
which may aid in B cell transformation. Human Papillomavirus is a com-
mon infectious agent found in epithelial cells and has been shown to pro-
mote malignant transformation in host cells through the overexpression
of oncoproteins E6 and E7. Hepatitis C Virus, whose life cycle is not fully
known yet, can also lead to carcinoma development in hepatocytes, and
HCV oncoproteins NS3, NS4A, and NS5B have been associated with onco-
genic roles. A novel genetic approach involving a CRISPR/Cas treatment
designed to dysregulate these viral oncogenes can be used to combat these
infections and their resulting carcinomas. While the long-term effects of
CRISPR treatments are still being researched, this gene therapy offers a
robust selection of potential treatments regarding long term diseases such
as oncogenic viral infections.
Figure 9: Bloom’s Taxonomy
1 Introduction
The human population is prone to cancer through a wide variety of variables;
in fact, 20% of which are caused by infectious agents such as bacteria, viruses,
and other pathogens.1 Oncogenic viruses alone are responsible for 12% of all
cancers diagnosed worldwide. More specifically, 80% of viral cancer cases occur
in the developing world today.2 The first oncovirus was discovered in 1964 when
researchers located the Epstein-Barr Virus in Burkitt’s Lymphoma cells using
electron microscopy.3 Since this turning point in viral oncology, six more viruses
have been confirmed to cause oncogenesis.
∗ Advised by: Everardo Hegewisch Solloa
1 [Hau09]
2 [Hau09]
3 [MHT17]
49 50
While infections from oncoviruses are common, they rarely lead to malig- reactivation of the virus.13 Recent advancements in molecular biology research
nancies and take a long period of time to transform host cells into an oncogenic have allowed researchers to explore gene-editing tools as potential therapeutics
state. For cancers to arise, these viral infections must be accompanied by chronic for oncogenic viruses. Clustered regularly interspaced short palindromic repeats
inflammation, environmental mutagens, or immunosuppression.4 Additionally, (CRISPR) technologies have now been considered for use in therapies to treat
oncogenic viruses do not follow a uniform path to oncogenesis since some of cancers, chronic viral infections, or genetic diseases.14 In situ, the CRISPR/Cas
the viruses are considered to be direct carcinogens, while others are indirect system is a bacterial and archaeal ”adaptive immune system” that recognizes
carcinogens. Direct carcinogens cause cancer cell transformation by expressing bacteriophage genomes complementary to a guide RNA (gRNA), which is bound
oncogenes, while indirect carcinogens cause chronic infection or inflammation, to a Cas protein. Once CRISPR/Cas has associated with its complementary
which leads to carcinogenesis.5 The seven known oncogenic viruses have been target sequence in the phage genome, it induces a double-stranded break in
characterized by the class of genetic material they possess (e.g., DNA or RNA). the complementary DNA strand, inhibiting the phage’s ability to replicate.15
The five known DNA oncoviruses include Epstein-Barr Virus (EBV), Hepatitis B CRISPR/Cas was discovered as a potential gene-editing tool in 2013 for its abil-
Virus (HBV), Human Herpesvirus-8 (HHV-8), Human Papillomavirus (HPV), ity to induce double-stranded breaks in a sequence-specific manner, resulting in
and Merkel Cell Polyomavirus (MCPyV). On the other hand, there are two either nonhomologous end-joining (NHEJ) or homology-directed recombination
oncogenic RNA viruses, which include Hepatitis C Virus (HCV) and Human (HDR).16 The ability to induce site-specific gene edits allows scientists to knock
T-Cell Lymphotropic Virus-1 (HTLV-1).6 out a gene or introduce a gene with higher efficiency. Not only has CRISPR/Cas
Oncogenic viruses can be transmitted via the exchange of bodily fluids and allowed scientists to further understand gene functions, but it also has the po-
direct skin to skin contact.7 Viral cancers only arise 15 to 40 years after primary tential of serving as a therapy via either the deletion or correction of a mutated
infection since chronic infection is a critical component of oncogenic transfor- gene.
mation. Oncoviruses can achieve chronic infection by switching from a lytic to a
latent stage after initially infecting host cells.8 Viruses in a lytic phase actively
induce the host cell to biosynthesize a wide variety of functional and structural
proteins to produce more viral progeny. Contrarily, viral latency does not in-
volve infectious virus production since, in a latent stage, viruses integrate a copy
of their genome into the host cell’s genome during viral replication and rely on
host cell replication for survival.9 The latent stage ensures that the virus exists
in host cells without detection from the immune system. It is often latency
that promotes oncogenesis since viruses in the lytic phase create a higher risk
of DNA damage detection that results in programmed cell death, inhibiting the
virus and host cell from replicating further. It is noted that cancer cells have
little to no evidence of viruses in the lytic stage.10
The most common treatments for viral cancers are currently chemother-
apy, such as lytic-inducing chemotherapies used for virus-related lymphomas,
and resection, which is the surgical removal of tumorous tissue.11 Additionally,
preventative treatments include vaccinations, which exist for some oncoviruses.
Other oncogenic viruses are treated using radiotherapy, immunotherapy, or an-
tiviral drugs prescribed on a case to case basis.12 However, these treatments
cannot guarantee complete eradication of infection and many cases result in
4 [MC10]
5 [MC10]
6 [MHT17]
7 [MSFP14]
8
Figure 1: Double-stranded breaks are induced by the Cas-gRNA complex
[MC10]
9 [CCY15] 13 [GP16]
10 [MC10] 14 [LdS19]
11 [MHT17]and [BDA+ 13] 15 [LdS19]
12 [CKQ19] 16 [LdS19]and [WHK15]
51 52
Currently, novel virus-targeted therapies are being researched for oncogenic The genome of Epstein-Barr Virus is a 172kb double-stranded DNA genome
viruses, which include new immunotherapies, cellular therapies, and antibody that encodes for lytic gene products and latent gene products. EBV infection
therapies.17 Gene therapies are treatments which consist of the application of is initiated when structural proteins gp350 and gp220 bind to CD21 receptors
zinc-finger nucleases (ZFN), transcription activator-like effector nuclease (TALEN), of B cells. EBV structural protein gp42 then completes viral fusion into B
and CRISPR to treat a disease. The concept of gene editing for repairing mu- cells by binding to the human leukocyte antigen class II receptor.24 Lytic gene
tated genes constitutes a new field of research that is being applied to various products are then expressed during the early stages of EBV infection and con-
diseases and disorders. Here I will propose a novel CRISPR/Cas based therapy sist of the early antigen (EA) complex, a complex of non-structural proteins,
for treating Epstein-Barr Virus, Human Papillomavirus, and Hepatitis C Virus. and the viral capsid antigen (VCA), which is composed of distinct structural
antigen complexes. Both the expression of EA and VCA are significantly re-
duced or not present after EBV transitions into a latent stage.25 EBV then
2 Epstein-Barr Virus (EBV) expresses latent proteins which fall under two main categories: nuclear antigens
(EBNA-1,2,3A,3B,3C,LP), and latent membrane proteins (LMP-1,2A,2B). All
The Epstein-Barr Virus is also known as human herpesvirus-4 and is one of the
nine proteins are expressed in both post-transplant lymphoproliferative disorder
eight human viruses in the herpesviridae family 2. Today, EBV has infected
(PTLD) cells and lymphoblastoid cell lines (LCL).26 EBV latency is categorized
90% of the world’s adults.18 However, the majority of EBV carriers maintain
into four stages (0-III) according to the proteins expressed during latency. B cell
lifelong, asymptomatic infections.19 EBV is a double-stranded DNA virus that is
immortalization is associated with latency stage III, which includes the expres-
transmitted primarily through saliva. After decades of research, it is confirmed
sion of all six nuclear antigen proteins and all three latent membrane proteins.
that EBV is able to infect a wide variety of cells that encompasses B cells,
Latency stage II encompasses the expression of proteins EBNA-1, LMP-1, LMP-
T cells, NK cells, glandular and squamous epithelial cells, and smooth muscle
2A, LMP-2B, and is most closely associated with Hodgkin Lymphoma. Latency
cells. However, the virus is most commonly known to have oncogenic potential
stage I only involves the expression of the EBNA-1 protein and most commonly
in B lymphocytes since it is able to create undetected DNA damage during a
occurs in Burkitt Lymphoma. The EBV genome persists in host cells with-
latent stage by inserting its genetic material into the host genome and then
out the expression of any viral proteins in latency stage 0, the latency stage
transforming B lymphocytes into proliferating lymphoblastoid cells through the
associated with non-dividing memory B cells.27
large amounts of oncogenic protein made during a long span of latency.20
LMP-1 is the main oncogenic protein involved in EBV induced oncogene-
sis. LMP-1 mimics the function of the CD40 receptor, which is a member of
2.1 Lifecycle and Genome Properties of EBV the tumor necrosis factor receptor (TNFR) superfamily, and activates down-
stream signaling pathways that are critical to the expression of anti-apoptotic
In its primary stage of infection, EBV is in a lytic phase and most commonly
proteins and differentiation in B cells.28 These signaling pathways include NF-
replicates in squamous epithelial cells and local lymphocytes.21 Cases of in-
B, MAPK/ERK, PI3K/AKT, Notch, and JAK/STAT. The most important
fectious mononucleosis can sometimes occur as a result of an abnormal EBV-
pathways for EBV induced oncogenesis are the PI3K/AKT and JAK/STAT
specific immune response.22 After EBV has established a local infection in a
pathways, which, once activated, will contribute to genomic instability, apopto-
lymph node, the secondary phase of infection begins when EBV spreads to
sis resistance, limitless replication, and tumor-promoting inflammation.29 Other
B lymphocytes, transforming them into virus-producing ’factories.’ Once in B
oncogenic proteins that have been shown to contribute to EBV related carci-
cells, EBV downregulates the expression of its growth-transforming gene so that
nomas include LMP-2A, EBNA-1, and EBNA-2 proteins. LMP-2A also con-
it can remain undetected in the host cell under a latent stage. Resting memory
tributes to oncogenesis by ensuring the survival of EBV infected cells through
B cells infected with latent EBV then continue to circulate as part of the B cell
the inhibition of TGF-1-induced apoptosis and activation of the Lyn/Syk sig-
pool. In a healthy host, EBV typically remains in a latent stage or is eradicated
naling pathway, which is a tyrosine kinase pathway essential for tumor survival.
by the immune system, yet in cases when EBV re-enters the lytic stage of in-
EBNA-1 is crucial for the maintenance and replication of the EBV genome and
fection at a mucosal surface, new viral particles are released and can go on to
also exhibits oncogenic behavior by suppressing the promyelocytic leukemia pro-
infect more cells.23
tein, a tumor suppressor protein responsible for regulating p53 activation. To-
17 [MHT17]
18 24 [SKJL10]
[MHT17]
19 25
[PGS+ 15] [YAM07]
20 26 [ESZM18]
[CCY15]
21 27 [DCH19]
[CCY15]
22 28 [PGS+ 15]
[PGS+ 15]
23 29 [MHT17]
[CCY15]
53 54
gether, EBNA-2 and EBNA-LP proteins are essential for transcription initiation sion of EBV treatments can be improved by using a more specialized approach,
of viral proteins LMP-1 and LMP-2A and cellular proteins MYC, CD21, and such as gene therapy.
CD23, all of which contribute to the transformation and immortalization of B
cells.30 The main oncogenic protein LMP-1 and oncogenic proteins LMP-2A, 2.3 A Potential CRISPR Treatment for EBV
EBNA-1, and EBNA-2 are essential components of EBV, which can be used as
potential targets in treatment. A CRISPR/Cas9 system designed to knock out EBV proteins LMP-1 And
EBNA-1 can be used as a treatment for persons with latent EBV infections
or EBV-induced B cell lymphomas since the Cas9 protein is able to induce
2.2 Epstein-Barr Virus Treatments
double-stranded breaks in DNA. DNA repair such as homologous recombination
Currently, the standard treatments for EBV associated carcinomas are systemic (HR) or nonhomologous end-joining (NHEJ) will then follow.36 The absence of
chemotherapy or radiotherapy. However, the majority of EBV associated malig- LMP-1 protein will inhibit viral ability to activate downstream signaling path-
nancies remain unaffected by these treatments, and there is currently no effective ways such as NF-B, MAPK/ERK, Notch, PI3K/AKT, and JAK/STAT, which,
therapeutic option for latent EBV tumors.31 A more increasingly used form when activated, all facilitate genomic instability, apoptosis resistance, limitless
of therapy for patients is combination therapy, which includes lytic-inducing replication, and tumor-promoting inflammation.37 This will decrease resistance
chemotherapy followed by the use of antiviral drugs. EBV viral replication to apoptosis and prevent unchecked cell proliferation in the host. Additionally,
is successfully suppressed by nucleoside analog antivirals, such as ganciclovir, with the absence of EBNA-1 protein, suppression of the promyelocytic leukemia
acyclovir, and famciclovir, when the virus is in a lytic stage. These antiviral protein, a tumor suppressor protein responsible for regulating p53 activation,
drugs rely on viral-encoded kinases, which are only expressed during a lytic will be prevented.38 This will result in the renovation of p53 functions and will
phase, to convert them to their active form where they are able to inhibit DNA most likely lead to programmed cell death.
polymerase of host cells, prevent viral DNA synthesis and kill tumor cells. A The main component of this treatment is a plasmid containing the CRISPR/Cas9
lack of activation, which is facilitated by latent EBV malignancies, will prevent gene, and DNA sequences for each target gene (LMP-1, EBNA-1) would need
antiviral drugs from successfully inhibiting viral replication. In contrast to the to be produced. This plasmid will be delivered into EBV infected cells via an
success shown by antiviral drugs, this approach to treating EBV related carcino- adenoviral vector, and will therefore also contain genes encoding for structural
mas also leads to a higher risk of viral transmission to surrounding healthy cells proteins of the viral vector and a packing signal to successfully package the plas-
since EBV can rapidly produce and spread new viruses during a lytic stage.32 mid into the adenoviral vector. Additionally, genes encoding for EBV structural
New treatments that are currently being researched as effective EBV-induced proteins gp350, gp220, and gp42 will be included in the plasmid to ensure that
carcinoma treatments include adoptive cell immunotherapy and the develop- the infection is directed towards B cells, the cells infected by EBV.39
ment of a vaccine to prevent primary EBV infection.33 Extensive research on Once in the host, these EBV-targeted viral vectors will execute fusion into B
the use of adoptive T cell therapy, which involves the use of EBV-specific cyto- cells. In the event that a healthy B cell obtains the plasmid, no genetic editing
toxic T cells (EBV-CTLs), on patients since 1995, is now confirming that this will take place since the EBV DNA needed to initiate replication of the plasmid
form of treatment can effectively prevent newly diagnosed and recurrent cases of is not present. However, once the plasmid has been successfully inserted into
EBV-PTLD.34 In addition to adoptive T cell therapy, a recombinant glycopro- EBV-infected B cells, transcription of the CRISPR engineered plasmid will be
tein gp350 vaccination was tested in healthy volunteers. While it has not been activated. After transcription, gRNAs complementary to oncogenes LMP-1 and
shown to prevent primary EBV infection yet, it has been shown to reduce cases EBNA-1 will associate with Cas9, and this Cas9-gRNA complex will bind to the
of symptomatic infectious mononucleosis. Like adoptive cell immunotherapy complementary strands of DNA encoding for LMP-1 and EBNA-1. The Cas9
for EBV, EBV vaccinations are aimed at preventing primary infection or EBV- protein will then create a double-stranded break at these sites. Nonhomologous
related malignancies through the induction of EBV-specific T cell responses. end joining (NHEJ) will follow, producing mutations that obstruct the function
These T cell responses can be achieved by stimulating T-cell mediated immunity of these genes, and the targeted oncogenes will be nonfunctional.
against viral antigens expressed during a latent stage.35 While these medical Western blotting can be used to validate these intended knockouts. The ini-
advancements have led to safer and more efficient EBV treatments, the preci- tial preparation for this process includes cell lysis and protein extraction. Cells
30 from an in vitro, CRISPR edited, EBV-positive B cell culture should be washed
[MHT17]
31 [MHT17] in a detergent, or buffer, such as phosphate-buffered saline, to induce cell dis-
32 [MHT17] 36 [LdS19]and [WHK15]
33 [KA14] 37 [MHT17]
34 [KA14] 38 [MHT17]
35 [KA14] 39 [LBZ+ 17]
55 56
background on blots and prevent antibodies from non specifically binding to the
membrane.43 The membrane will detect antibodies corresponding to EBV pro-
teins LMP-1 and EBNA-1 via an enzyme, such as horseradish peroxide, which
will produce a signal based on the position of the target proteins. Finally, this
signal will be captured on a film, quantifying the presence of target proteins.44
The western blots from these CRISPR edited cells can then be compared to
EBV-positive B cells, which did not receive treatment. After comparison, the
intended knockouts can be considered successful if the demarcations of LMP-1
and EBNA-1 were faint or nonexistent on the CRISPR edited cell western blots
compared to those of untreated EBV cells.
Trials for this treatment should first take place in vitro, using transgenic,
EBV-positive mice cells, progress to in vivo mice trials after knockouts have
proven successful, and then take place in EBV-positive human B cells in vitro
before moving to human trials. Successful in vivo EBV trials centered around
studying the role of EBV protein LMP-1 in lymphomagenesis in mice models
have already been done by [ZKY+ 12].45 They generated a Rosa26 allele, which
allowed for LMP-1 expression through the excision of a transcriptional/ trans-
lational STOP cassette via Cre/loxP-mediated recombination (LMP1flSTOP).
LMP-1 expression was then induced in B cells from the pre-B cell stage by cross-
ing the LMP1flSTOP mice to CD19-cre mice.46 These B cells were then used
to study various LMP-1 interactions and functions. A similar approach can be
Figure 2: EBV plasmid containing Cas9 gene, oncogenes LMP-1 and EBNA-1, taken when modeling EBV-positive B cell lines and EBV-infected mice for in
structural adenovirus proteins, a packaging signal, and genes encoding for EBV vitro and in vivo trials, respectively. Furthermore, trials testing the efficacy of a
proteins gp350, gp220, and gp42 CRISPR-Cas9 treatment have already been carried out by [WQ14] in vitro us-
ing cells from Burkitt’s lymphoma patients, and have shown to be successful.47
EBV-positive cells from the Raji cell line, the first established long-term culture
ruption. Protein extraction should then take place under a cold temperature from Burkitt’s lymphoma patients, were used for testing. This EBV-targeted
with protease inhibitors so that the denaturing of proteins is prevented.40 Af- CRISPR/Cas system included gRNA for EBNA1-7, and knockouts were vali-
ter protein extraction, the concentration of proteins should be measured using dated using western blotting.48 Results from these trials showed a significant
a spectrophotometer, and the protein sample should be diluted with a loading decline in cell proliferation and viral load and showed restoration in apoptotic
buffer containing glycerol so that the sample will easily sink into wells containing pathways. The research done by both Zhang et al. and Wang and Quake can
gel. The sample should then be heated since heating will denature the sample be referred to when considering appropriate testing procedures and models for
and give proteins a negative charge, a key component for movement in an elec- the potential CRISPR-Cas9 treatment.
tric field when a voltage is applied.41 After this sample preparation, the protein
sample will go through gel electrophoresis, which involves both a stacking and a
separating agarose gel. The stacking agarose gel is above the separating gel and 3 Human Papillomavirus (HPV)
has an acidic pH of 6.8, which will separate the proteins into sharply defined
bands. The separating agarose gel has a basic pH of 8.8, allowing for narrower Human Papillomaviruses belong to the Papillomaviridae family, a large family
gel pores, which will separate proteins by size since smaller proteins will sink of double-stranded DNA viruses.49 HPV is the world’s most commonly trans-
faster.42 After a voltage has been applied to the gel, the separated protein mitted sexual disease and primarily infects epithelial cells through skin to skin
mixture will be transferred to either a nitrocellulose or polyvinylidene fluoride 43 [MY12]
membrane via electrophoretic transfer. A blocking solution containing either 5% 44 [MY12]
45 [ZKY+ 12]
BSA or TBST diluted, nonfat dried milk should then be added to reduce the 46 [ZKY+ 12]
40 47 [WHK15]and [WQ14]
[MY12]
41 48 [WQ14]
[MY12]
42 49 [GS16]and [HM17]
[MY12]
57 58
or mucosa to mucosa contact.50 Over 200 types of HPV are known to exist and the host genome instead of existing as an episome.59
can be further classified into five genera which include alpha-papillomaviruses, The HPV genetic material exists as an 8kb non-enveloped, circular DNA
beta-papillomaviruses, gamma-papillomaviruses, mu-papillomaviruses, and nu- genome.60 The HPV genome is composed of a long, non-coding control region,
papillomavirus.51 Skin infections are caused by beta, gamma, mu, and nu- six early-transcribed genes (E1,2,4,5,6,7), which encode for the non-structural
HPVs, while mucosa infections are caused by alpha-HPVs.52 The most common proteins, and two late-transcribed genes (L1,2) which encode for the structural
visible outcome of chronic HPV infection is the appearance of benign epithelial viral capsid proteins.61 Early proteins E1,2 and 4 are responsible for the gene
warts around the site of primary infection. Healthy individuals that are infected regulation, replication, and pathogenesis of the virus, while late proteins L1
with HPV rarely show symptoms, and few actually develop disfiguring warts. (major viral capsid protein) and L2 (minor viral capsid protein) are involved in
Still, the lifetime risk of a sexually transmitted HPV infection is 50%, and some assembling virus-like particles. In fully matured virions infecting new cells, L1
cases of chronic and/or repetitive HPV infection can lead to the development is also the protein responsible for facilitating entry into host cells by binding to
of malignancies.53 In terms of malignancies, HPV types can also be categorized the widely expressed, host cell surface receptor HSPG.62 HPV proteins E1 and
into low risk and high risk types. HPV types 6 and 11 are the most common low E2 facilitate viral DNA replication and regulate early transcription, and E4 is
risk types and cause external anogenital warts in 90% of infections.54 All high thought to aid in viral escape from the cornified layer of the epithelium.63
risk types of HPV are classified as genera, with the most common types being HPV genes E5,6 and 7 are shown to possess oncogenic properties with genes
HPV-16,18,31,33,52,58. Furthermore, HPV-16 and 18 are most commonly asso- E6 and E7 playing the main role in transforming healthy host cells into malig-
ciated with malignancies. Although HPV-associated cancers are rare compared nant growths and gene E5 assisting in oncogenesis by promoting cell prolifera-
to other malignancies, they are still responsible for 99% of cervical cancer, 85% tion through the activation of tyrosine kinase receptors EGF and PDGF.64 E6
of anal cancer, and 50% of genital-associated cancers.55 and E7 proteins are the main difference between low risk and high risk HPV
infections since they have been confirmed to function as oncoproteins in high
3.1 Lifecycle and Genome Properties of HPV risk infections and not low risk infections.65 E7 is responsible for binding to
cellular factors from the retinoblastoma (Rb) family, such as p105 (RB), p107,
While the majority of viruses can produce progeny viruses after infecting a tar- and p130, and degrading them. While this is done in all HPV infections, E7
get cell, HPVs are only able to synthesize new virions after target cells have binds extensively to Rb cellular factors in high risk infections.66 Similarly, in
undergone mitosis, and infected daughter cells have differentiated. In healthy high risk infections, E6 is able to efficiently bind to tumor suppressor p53 and
epithelia, basal cells are the only proliferating cells and are exposed as a result degrade it via the ubiquitination-mediated pathway. On the other hand, E6 is
of micro-wounds. These cells in the basal layer of stratified squamous epithelia unable to easily bind to p53 and inactivate it in low risk infections.67
serve as the site of primary infection for HPV.56 After primary infection, an epi- Evidence from decades of cervical cancer research suggests that HPV has
some is established as the HPV genome and does not encode for the polymerases certain growth advantages when its genome is integrated into the genome of the
and enzymes responsible for viral replication. As a result, HPV relies entirely host cell as opposed to remaining as an episome. Interestingly, HPV protein
on host cells for replication. HPVs full life cycle matches the time it takes for E2, which is a transcriptional repressor for oncoproteins E6 and E7, is often the
epithelial cells to fully mature, a cycle which lasts 2-3 three weeks.57 During this site of integration. This disruption of E2 expression then results in an increased
time, suprabasal cells continue to follow a normal life cycle while a subset of cells expression of E6 and E7. In high risk HPV infections, a combination of increased
containing viral episomes re-enter the DNA synthesis phase (S phase) and en- expression in both main oncoproteins and the inactivation of tumor suppressor
gage in amplification, the process by which HPV genomes are replicated.58 Over pathways Rb and p53 lead to large amounts of genomic instability in the host
time, frequently recurring HPV infections lead to the accumulation of cellular and ultimately, an increased progression in malignancies.68 These advancements
mutations. After several decades of failure to clear persistent HPV infections, in the understanding of the HPV genome can serve as a foundation for new
malignancies develop from the extensive accumulation of these mutations. In HPV-targeted treatments.
cancerous cells, the HPV genome is also commonly found to be integrated into 59 [ML10]
50 60 [MHT17]
[GS16]
51 61 [GS16]
[MHT17]and [HM17]
52 62 [HBR+ 10]
[MHT17]
53 63 [HM17]and [ZB06]
[MHT17]
54 64 [HM17]and [ML10]
[BFM17]
55 65 [GS16]
[GS16]
56 66 [ML10]
[ML10]
57 67 [GS16]
[MHT17]
58 68 [GS16]
[ML10]
59 60
3.2 Human Papillomavirus Treatments targeted sites since HPV is also a double-stranded DNA virus. This treatment
will be inserted into HPV-infected cells via an adenoviral vector.
Several preventative HPV vaccinations have been created in the twenty-first
The plasmid containing the CRISPR-Cas system will include the CRISPR-
century. Gardasil, a quadrivalent vaccine for prevention against HPV types
Cas9 gene, DNA sequences for HPV oncogenes E6 and E7, genes encoding for
6, 11, 16 and 18 was licensed in 2006 for use in the United States. In 2009,
structural proteins of the adenovirus, a packaging signal, and the gene for HPV
Cervarix, a bivalent vaccination against HPV 16, and 18 was approved for use
structural viral capsid protein L1; the protein responsible for binding to basal
in the United States.69 The most recent vaccination for HPV, a 9-valent vaccine
layer epithelial cells. Once the plasmid is successfully inserted into HPV-positive
(9vHPV) which prevents against HPV types 6, 11, 16, 18, 31, 33, 45, 52, and
cells and transcription has occurred, gRNAs complementary to HPV oncogenes
58, was licensed in 2014 for use in the United States, and 2015, for use in
E6 and E7 will associate with the Cas9 proteins. This Cas9-gRNA complex will
Europe. Today, 9vHPV is the only HPV vaccination in the United States and
then bind to the E6 and E7 sequences complementary to it, and Cas9 will induce
is also licensed for use in Canada, Australia, Chile, Hong-Kong, Ecuador, South
double-stranded breaks in both oncogenes. Afterward, NHEJ will occur, and the
Korea, and New Zealand.70 All preventative vaccines that have been created
mutations originating from this repair process will obstruct proper expression of
for HPV contain a synthetic recombinant L1 major capsid protein base. Still,
HPV oncogenes E6 and E7, inhibiting their functions. These knockouts will also
while all HPV vaccines have proven effective in preventing HPV, the L1 base
be validated using western blots. Westerns blots of in vitro, CRISPR edited,
restricts these vaccinations to HPV type-specific responses and does not offer a
HPV-positive cells should be compared with western blots of untreated HPV-
broader spectrum of defenses against all human papillomaviruses.71
infected cells to confirm a significant decrease of E6 and E7 expression.
New treatments that are being researched for HPV prevention and HPV
associated carcinomas include an L2 based vaccination and therapeutic vacci-
nations. The L2 minor capsid protein possesses homology across HPV types,
which would provide broader coverage of HPVs if used as the base of a vac-
cine.72 Therapeutic vaccinations are another treatment option for HPV induced
malignancies or pre-malignancies. These vaccines are being aimed to generate
antigen-specific, cellular-mediated immunities instead of the humoral immuni-
ties which can be gained from preventative HPV vaccinations. Therapeutic
vaccines are designed to induce CD8+ cytotoxic T cells and CD4+ helper T
cells to target epithelial cells that contain E6 and E7, the two main viral onco-
proteins.73 A similar precision-based approach, which can be applied directly
to infected cells, can be achieved through the use of new tools involved in gene
editing.
3.3 A Potential CRISPR Treatment for HPV

An engineered CRISPR-Cas9 system targeting main HPV oncogenes E6 and E7
can be used to treat HPV-related carcinomas. When knocked out, the absence of
E6 will inhibit the degradation of tumor suppressor p53.74 An E7 knock out will
prevent degradation of cellular factors from the Rb family such as RB, p107,
and p130.75 By disabling the expression of both oncogenes, host cell tumor
suppressors will be able to function and will detect tumorigenesis, leading to
programmed cell death. Like EBV, the Cas9 protein will be used to cleave
Figure 3: HPV plasmid containing Cas9 gene, oncogenes E6 and E7, structural
69 [MHT17]
70
adenovirus proteins, a packaging signal, and HPV gene L1
[SOC+ 17]
71 [MHT17]
72 [MHT17] Like the proposed treatment for EBV, trials for this treatment should take
73 [MHT17] place in humanized mice models both in vitro and in vivo before using in vitro
74 [GS16]
75
human cell lines, and eventually progressing to in vivo human trials. [BPHD+ 12]
[ML10]
were able to achieve humanized mice models of HPV-associated pathology via
61 62
the expression of HPV protein E7.76 Artificial human skin, which was pre- 4.1 Lifecycle and Genome Properties of HCV
pared using primary keratinocytes engineered to express E7, was transplanted
Like other oncogenic viruses such as EBV and HPV, HCV infections progress
into nude mice. These E7 genes were obtained from HPV strains 5, 10, and
into malignancies after a long period of time, taking up to 20-40 years to fully
16 by using PCR and specific primers from plasmids pBP-5E7, pcDNA3-10E7,
progress into hepatocellular carcinoma.85 On the other hand, while HPV and
and pBP-16E7. Once transplanted into mice, these E7 transplants were sta-
EBV are able to integrate their genetic material into the host genome, HCV
bly maintained for six months and began to promote anogenital warts after
is a single-stranded RNA virus that is incapable of genome integration. As
this time period, a significant symptom of oncogenic strains of HPV.77 A sim-
a result, HCV spends the majority of its life cycle in the cytoplasm of host
ilar approach, based on the expression of both E6 and E7 oncogenes, can be
cells and is assumed to cause carcinogenesis through indirect mechanisms, even
taken when mimicking HPV in mouse models for this potential CRISPR treat-
though some components of its life cycle remain unknown.86 Upon primary
ment. Additionally, [KKG+ 14] were able to achieve successful in vitro results
infection, HCV attaches to host cells via target cell receptors such as CD81, SR-
from a CRISPR-Cas9 system designed to knock out HPV oncoproteins E6 and
B1, LDL-R, EGFR, and EphA2. It then uses a clathrin-mediated endocytosis
E7.78 This treatment was applied to HeLa and SiHa cell cultures, which were
process to enter the host cell and release its RNA genome in the cytoplasm.87
grown in Dulbecco’s modified Eagle medium and supplemented with 10% fetal
After successful translation of the HCV genome, new virions undergo assembly
bovine serum. The knockouts were verified using western blotting, and results
and maturation in an endoplasmic reticulum compartment where they are also
showed reactivation of tumor suppressors such as p53 and cellular factors from
surrounded by endogenous lipoproteins, which are believed to aid in immune
the retinoblastoma family.79 Both these studies can be used as procedure and
escape. New virions are then believed to exit cells via exocytosis.88 Chronic
analysis references for trials of the potential CRISPR treatment.
damage to hepatocytes as a direct result of chronic HCV infection induces the
release of inflammatory and fibrotic mediators such as reactive oxygen species
4 Hepatitis C Virus (HCV) (ROS), cell death signals, hedgehog ligands, and nucleotides.89 This creates
genomic instability, making host cell genomes susceptible to modification and
Hepatitis C Virus is a single-stranded RNA virus that belongs to the Hepacivirus transformation by the expressions of HCV proteins believed to play roles in
genus and is a member of the Flaviviridae family.80 HCV is primarily trans- oncogenesis.
mitted via percutaneous or mucosal contact with infected blood, while other The HCV genome is an enveloped, single-stranded RNA genome, measuring
routes of transmission include high risk sexual activities and contact with other approximately 9.6 kilo-bases, and encodes for ten proteins. These proteins in-
infected body fluids.81 HCV can cause both acute and chronic liver infections clude three structural proteins (Core, E1, E2) and seven non-structural proteins
but has shown to progress to chronic infections in 75-85% of infected persons. (p7 viroporin, NS2, NS3, NS4A, NS4B, NS5A, NS5B). HCV structural proteins
An HCV infection is considered chronic if, six months after primary infection, Core, E1, and E2, as well as non-structural proteins p7 viroporin and NS2, are
HCV RNA persists in the blood. This type of infection is commonly induced by early expressed proteins involved in virus assembly and release.90 It is also be-
HCV because it is a frequently replicating virus with an RNA genome that is lieved that E2 is the protein responsible for binding to HCV entry factors on
highly prone to replication errors, allowing for long-term evasion of the host im- host cell surfaces.91 HCV non-structural proteins NS3 and NS4A make up the
mune system. As a result, innate immune responses in the host are significantly NS3-4A serine protease complex responsible for cleavage at four different sites
delayed, and a widespread HCV infection is often established before adaptive of the HCV non-structural polyprotein precursor, which include NS3/NS4A
immune cell responses are activated.82 Today, 170 million people are chronically (self-cleavage), NS4A/NS4B, NS4B/NS5A, and NS5A/NS5B.92 Non-structural
infected with HCV worldwide, and the annual rate of newly diagnosed cases is protein NS4B is a membrane-associated protein that provides mediation be-
4 million per year.83 Due to its high potential to cause chronic infections, HCV tween virus-host interactions. Non-structural proteins NS5A and NS5B are
is the leading cause of cirrhosis and is considered an indirect carcinogen for both involved in HCV RNA replication; NS5A is a zinc-binding, proline-rich,
hepatocellular carcinoma, which is mainly prevalent in developed countries.84 hydrophilic, phosphoprotein, while NS5B is an RNA dependent RNA poly-
76 85
[BPHD+ 12] [GH15]
77 86
[BPHD+ 12] [BTBZ09]
78 87
[KKG+ 14] [Dus16]
79 88
[KKG+ 14] [Dus16]
80 89
[MHT17]and [CM06]and [SEKI+ 16] [GH15]
81 90 [Dus16]
[MHT17]and [CM06]
82 91 [PE12]
[Dus17]
83 92 [Lin70]
[MHT17]
84 [MHT17]and [SEKI+ 16]
63 64
merase.93 this form of therapy more tolerable for patients. When taken together or in com-
HCV is primarily known to promote carcinogenesis through indirect mecha- bination with PEG-IFN plus RBV, simeprevir and sofosbuvir raised SVR rates
nisms, but direct oncogenic activity is also a possibility that is currently being to more than 90%.101 Since this initial breakthrough, many other DAAs, many
researched. Indirect carcinomic mechanisms include the release of profibrogenic of which are oral, have become available for combating a wide variety of HCV
cytokines and chemokines such as TGF-, which has shown to tumor suppressor genotypes and stages of liver infection. Today, different DAAs target different
properties in healthy cells and fibrogenic activity under chronic inflammation.94 HCV proteins, which include the NS3/4A protease (”-previr” DAAs), NS5B
While HCV has not explicitly shown direct carcinogenic properties, ongoing re- polymerase (”-buvir” DAAs), and NS5A inhibitors (”-asvir” DAAs).102 How-
search suggests that structural protein Core and non-structural proteins NS3, ever, while DAAs have provided many advancements regarding the efficiency
NS4A, NS4B, NS5A, and NS5B could possess oncogenic qualities since they of HCV treatments through high rates of SVR, outcomes of these treatments
have shown to promote oncogenesis via direct interaction with cellular factors have been limited to reducing the risk of HCV-induced hepatocellular carcinoma
involved in host cell cycle progression, apoptosis, DNA replication, DNA re- instead of eliminating it.103
pair, and angiogenesis.95 Core, NS4B, and NS5A proteins have been shown
to activate the Wnt/-catenin signaling pathway, a pathway associated with tu- 4.3 A Potential CRISPR Treatment for HCV
mor cell growth, metastasis, and hepatocellular carcinoma recurrence, in Huh7
cells, which are from a well-differentiated hepatocyte derived carcinoma line.96 Chronic HCV infections and HCV-induced malignancies can be treated using
NS3 and NS4A proteins have been shown to inhibit DNA repair processes and a CRISPR-Cas13 system targeting HCV genes NS3, NS4A, and NS5B, which
interact with ATM, a protein in host cells responsible for DNA damage detec- have been shown to directly promote cell transformation and oncogenesis. The
tion.97 Additionally, evidence suggests that NS5B protein binds to and inhibits Cas13 protein is capable of inducing single-stranded breaks in RNA, making
host cell tumor suppressor protein Rb, allowing transformed cells to enter S it compatible with HCV since it is a single-stranded RNA virus.104 Without
phase undetected. While, NS5A blocks the activation of caspase-3, preventing NS3 and NS4A expression, the HCV NS3/4A serine protease complex will be
TNF--mediated apoptosis, and NS3, NS4A, and NS5B proteins facilitate the prevented from interacting with host cell protein ATM, a cellular factor involved
relocalization of tumor suppressor p53 from the host cell nucleus to the cy- in DNA damage detection.105 This will allow for a better immune response to
toplasm; an occurrence which interferes with p53-induced apoptosis.98 HCV HCV since DNA damage is a significant outcome of chronic HCV infection,
proteins, which have shown potential oncogenic properties, serve as the basis of which can exist and progress unchecked when NS3/4A is expressed, leading to
research for novel HCV-targeted treatments. malignancies. An NSB knockout will prevent HCV NS5B protein from binding
to and inhibiting host cell tumor suppressor protein Rb, allowing for proper
tumorigenesis detection.106 Together, the absence of NS3, NS4A, and NS5B
4.2 Hepatitis C Virus Treatments
will also prevent the relocalization of tumor suppressor p53 from the host cell
HCV was historically treated with pegylated-interferon (PEG-IFN) alpha plus nucleus to the cytoplasm.107 This will allow p53 to induce apoptosis without
ribavirin (RBV) for 24 to 48 weeks and was designed to create a sustained viro- interference or obstruction.
logical response (SVR). However, this treatment only produced SVR in 40-50% This treatment will consist of a CRISPR-Cas13 engineered plasmid, which
of patients and frequently caused side effects such as hemolytic anemia, flu-like will be inserted into host cells via an adenoviral vector. The plasmid will contain
symptoms, and psychiatric disturbances after treatment.99 In 2011, a combi- the CRISPR/Cas13 gene, HCV genes NS3, NS4A, and NS5B, genes encoding for
nation of newly approved direct-activating antivirals (DAAs), boceprevir and structural proteins of the viral vector, a packing signal, and the gene encoding
telaprevir, and PEG-IFN plus RBV were able to increase SVR to almost 70% for HCV protein E2, which is needed for HCV entry into host cells. After the
but required patients to take heavy dosages and confined them to strict dietary plasmid has successfully entered HCV-infected cells, transcription of the plasmid
requirements.100 DAAs simeprevir and sofosbuvir were released in 2013 and will occur. The gRNAs consisting of complementary sequences to HCV genes
took a new approach to HCV treatment. They were the first oral daily treat- NS3, NS4a, and NS5B will then associate to Cas13 protein and form a Cas13-
ments requiring one daily dose and were accompanied by no side effects, making gRNA complex. Following this process, the Cas13-gRNA complex will bind to
93
the respective HCV oncogene sequences, and Cas13 will induce a single-strand
[Dus16]
94 [MHT17] 101 [KAS17]
95 [MHT17] 102 [KAS17]
96 [MHT17] 103 [GH15]
97 104
[MHT17] [LdS19]and [SEKI+ 16]
98 [MHT17] 105 [MHT17]
99 [KAS17] 106 [MHT17]
100 [KAS17] 107 [MHT17]
65 66
break at these sites of the HCV RNA genome, disabling the expression and and ONCL.109 These findings must be considered when developing a future HCV
function of these genes. Like the potential CRISPR treatments for EBV and mouse model for testing potential treatments aimed at eliminating the risk of
HPV, the intended knockouts in this treatment can be validated using western HCV-associated carcinomas. On the other hand, [PSR+ 15] were able to carry
blots. A western blot of successful knockouts for NS3, NS4A, and NS5B should out a successful in vitro trial of a CRISPR-Cas treatment, which prevented HCV
reveal a significant decrease in the quantity of these proteins in vitro, compared replication.110 [PSR+ 15] constructed vectors encoding for a Francisella Novicida
to HCV-positive cells that did not receive the CRISPR treatment. Cas9 endonuclease (FnCas9), which has been shown to cleave RNA strands in
bacteria and archaea, and the HCV’5 untranslated region, which has been shown
to be involved in translation of the viral polyproteins and replication of the viral
RNA. The vectors were then inserted into HCV-infected human Huh7 cells in
vitro, and knockouts were verified using PCR afterward.111 The results showed
a significant decrease in polyprotein translation and viral replication. While a
mouse model might not be able to properly simulate a high risk HCV infection,
CRISPR trials have been done using human Huh7 cell cultures, an approach
that should be referenced when testing this potential CRISPR-Cas13 treatment.
5 Conclusion
Oncogenic viruses are an increasing cause of cancer worldwide, with 12% of all
cancers resulting from these viral infections today.112 Epstein-Barr Virus, Hu-
man Papillomavirus, and Hepatitis C Virus all play direct roles in carcinogenesis.
EBV promotes B cell transformation by integrating its genome into the genome
of the host cell, and expresses oncoproteins LMP-1, LMP-2A, EBNA-1, and
EBNA-2. LMP-1 is the main mechanism of oncogenesis since it activates down-
stream signaling pathways, such as NF-B, MAPK/ERK, PI3K/AKT, Notch,
and JAK/STAT, which all promote genomic instability, inflammation, resis-
tance to apoptosis, and unchecked proliferation in host cells.113 In the majority
of HPV-associated malignancies, infected host cells are also found to have an
HPV genome integrated into their own. This promotes the overexpression of
Figure 4: HCV plasmid containing Cas13 gene, oncogenes NS3, NS4A, and
HPV oncoproteins E6 and E7, which degrade cellular tumor suppressors such
NS5B, structural adenovirus proteins, a packaging signal, and HCV gene E2
as p53 and members of the Rb family (RB, p107, and p130).114 HCV has been
known to indirectly cause malignancies via chronic infection and inflammation,
Unlike HPV and EBV, which can infect a variety of species, HCV has only
which leads to tumor-promoting DNA damage. However, recent studies suggest
been shown to infect humans and chimpanzees. As a result, no successful mouse
that HCV proteins NS3, NS4A, NS4B, NS5A, and NS5B could potentially play
models which mimic the rapid HCV replication found in human malignancies
direct roles in tumorigenesis by activating the Wnt/-catenin pathway, inhibiting
have been achieved yet, making it impractical to effectively execute preliminary
DNA repair processes, preventing Rb functions, and relocalizing p53 from the
trials of this potential CRISPR-Cas13 treatment in mice. However, [DHR+ 11]
host nucleus to the cytoplasm.115 A thorough understanding of oncoproteins in
were able to induce a slight HCV infection in Rosa26-fluc mice.108 Recombi-
EBV, HPV, and HCV is critical for the development of efficient, viru-targeted
nant adenoviruses encoding for human cell surface receptors CD81, SCARB1,
treatments.
CLDN1, and occludin (OCLN) were inserted into murine livers, making the mice
The research field for oncogenic virus treatments is continuously evolving
susceptible to HCV since it is able to bind to these receptors. While HCV did
109 [DHR+ 11]
not express all proteins that would be expressed in a human HCV infection, and
110 [PSR+ 15]
it did not replicate or spread with the same speed as it would in human hepato- 111 [PSR+ 15]
cytes, results from this experiment show that HCV infections can be induced in 112 [Hau09]
mice if they are able to express a minimum of human cell surface receptors CD81 113 [MHT17]
114 [GS16]and [ML10]
108 [DHR+ 11] 115 [MHT17]
67 68
because of more precise therapies such as CRISPR. In the past, chemotherapy [BFM17] Pina Brianti, Eduardo De Flammineis, and Santo Raffaele Mercuri.
was the most common treatment for patients, but it often resulted in an ineffi- Review of hpv-related diseases and cancers. The New Microbiolog-
cient or incomplete elimination of virus-related cancers. New treatment options ica, 2017.
involving antiviral drugs have been able to slow the process of tumor develop-
ment, while preventative vaccinations have helped to lower the risk of infection [BPHD+ 12] Agueda Buitrago-Perez, Mariam Hachimi, Marta Duenas, Belen
that could result in malignancies. However, these options have not been able to Lloveras, Almudena Santos, Almudena Holguin, Blanca Duarte,
eliminate the risk of oncogenesis because they do not target the specific mech- Juan Luis Santiago, Baki Akgul, Jose L. Rodriguez-Peralto, Alan
anisms directly related to cell transformation and lack the precision to do so. Storey, Catalina Ribas, Fernando Larcher, Marcela del Rio, Je-
Novel gene therapies are able to provide this level of precision, and a CRISPR sus M. Paramio, and Ramon Garcia-Escudero. A humanized mouse
based treatment designed to knock out oncogenic proteins can facilitate a safer model of hpv-associated pathology driven by e7 expression. PLOS
and more efficient treatment compared to current treatments. ONE, 2012.
Results for trials of this precise, virus-targeted, CRISPR treatment have [BTBZ09] Birke Bartosch, Robert Thimme, Hubert E Blum, and Fabien
shown that this approach effectively induces the renovation of apoptotic path- Zoulim. Hepatitis c virus-induced hepatocarcinogenesis. Journal
ways and cell death in infected cells. These natural methods of tumor suppres- of Hepatology, 2009.
sion, induced by CRISPR treatments, might also allow for faster elimination
of malignancies, therefore providing a more cost-efficient option for patients [CCY15] Qingqing Cai, Kailin Chen, and Ken H. Young. Epstein-barr virus-
and treatment industries. However, more research regarding the likelihood of positive t/nk-cell lymphoproliferative disorders. Experimental and
CRISPR-induced off-target edits and their effects must still be done. While Molecular Medicine, 2015.
CRISPR is able to precisely edit genes to correct their functions, it can also
lead to off-target mutations that can dysregulate the expression of other, nor- [CKQ19] Jungang Chen, Samantha Kendrick, and Zhiqiang Qin. Mechanis-
mal genes. This could result in secondary diseases, or it could potentially worsen tic insights into chemoresistance mediated by oncogenic viruses in
the conditions of the existing disease. Both these desirable and counterproduc- lymphomas. Viruses, 2019.
tive outcomes of CRISPR treatments must be fully considered before clinical [CM06] Stephen L. Chen and Timothy R. Morgan. The natural history of
use. hepatitis c virus (hcv) infection. International Journal of Medical
Additionally, while both in vitro and in vivo trials have produced desirable Sciences, 2006.
outcomes, the long term effects of CRISPR treatments are still unknown. The
chances of viral infection recurrences and carcinoma regeneration must also be [DCH19] James P Dugan, Carrie B Coleman, and Bradley Haverkos. Op-
calculated. Current treatments are unable to fully eliminate the chances of these portunities to target the life cycle of epstein-barr virus (ebv) in
recurrences, but the efficacy of CRISPR in these aspects of treatment have not ebv-associated lymphoproliferative disorders. Frontiers in Oncol-
been fully observed yet because it has existed as a gene-editing tool for less than ogy, 2019.
a decade. Virus-targeted CRISPR trials carried out on animal models, or human
subjects will require years of observation after initial treatments to ensure that [DHR+ 11] Marcus Dorner, Joshua A. Horwitz, Justin B. Robbins, Walter T.
recurrences of viral infections and carcinomas do not occur. Furthermore, the Barry, Qian Feng, Kathy Mu, Christopher T. Jones, John W.
ethical consequences of inaccurate gene edits and the manipulation of nature Schoggins, Maria Teresa Catanese, Dennis R. Burton, Mansun
should be fully explored before using CRISPR in a clinical setting. Nevertheless, Law, Charles M. Rice, and Alexander Ploss. A genetically hu-
this approach to treating oncogenic viruses and a wide variety of other genetic manized mouse model for hepatitis c virus infection. Nature, 2011.
diseases will be able to produce precise results in future treatment settings. [Dus16] L. B. Dustin. Hepatitis c virus: Life cycle in cells, infection and
host response, and analysis of molecular markers influencing the
outcome of infection and response to therapy. Clinical Microbiology
References and Infection : the Official Publication of the European Society of
[BDA+ 13] Ulas Darda Bayraktar, Luis A. Diaz, Brittany Ashlock, Ngoc Clinical Microbiology and Infectious Diseases, 2016.
Toomey, Lisa Cabral, Soley Bayraktar, Denise Pereira, Dirk P.
[Dus17] Lynn B. Dustin. Innate and adaptive immune responses in chronic
Dittmer, and Juan Carlos Ramos. Zidovudine-based lytic-inducing
hcv infection. Current Drug Targets, 2017.
chemotherapy for epstein–barr virus-related lymphomas. HHS Au-
thor Manuscripts, 2013.
69 70
[ESZM18] Ahmed El-Sharkawy, Lobna Al Zaidan, and Ahmed Malki. [LdS19] Alexandra Loureiro and Gabriela Jorge da Silva. Crispr-cas: Con-
Epstein-barr virus-associated malignancies: Roles of viral onco- verting a bacterial defence mechanism into a state-of-the-art ge-
proteins in carcinogenesis. Frontiers in Oncology, 2018. netic manipulation tool. Antibiotics (Basel, Switzerland), 2019.
[GH15] Nicolas Goossens and Yujin Hoshida. Hepatitis c virus-induced [Lin70] Chao Lin. Hcv ns3-4a serine protease. Hepatitis C Viruses:
hepatocellular carcinoma. Clinical and Molecular Hepatology, 2015. Genomes and Molecular Biology, 1970.
[GP16] Stevan A. Gonzalez and Robert P. Perrillo. Hepatitis b virus reac- [MC10] Patrick S. Moore and Yuan Chang. Why do viruses cause cancer?
tivation in the setting of cancer chemotherapy and other immuno- highlights of the first century of human tumour virology. Nature
suppressive drug therapy. Clinical Infectious Diseases : an Official Reviews. Cancer, 2010.
Publication of the Infectious Diseases Society of America, 2016.
[MHT17] Uyen Ngoc Mui, Christopher T. Haley, and Stephen K. Tyring.
[GS16] Ge Gao and David I. Smith. Human papillomavirus and the de- Viral oncology: Molecular biology and pathogenesis. Journal of
velopment of different cancers. Cytogenetic and Genome Research, Clinical Medicine, 2017.
2016.
[ML10] Cary A. Moody and Laimonis A. Laimins. Human papillomavirus
[Hau09] Harald Zur Hausen. The search for infectious causes of human oncoproteins: pathways to transformation. Nature Reviews. Can-
cancers: where and why. Virology, 2009. cer, 2010.
[HBR+ 10] Caroline A.J. Horvath, Gaëlle A.V. Boulet, Virginie M. Renoux, [MSFP14] Abigail Morales-Sánchez and Ezequiel M. Fuentes-Pananá. Human
Philippe O. Delvenne, and John-Paul J. Bogers. Mechanisms of cell viruses and cancer. Viruses, 2014.
entry by human papillomaviruses: an overview. Virology Journal,
2010. [MY12] Tahrin Mahmood and Ping-Chang Yang. Western blot: Technique,
theory, and trouble shooting. North American Journal of Medical
[HM17] Mallory E. Harden and Karl Munger. Human papillomavirus Sciences, 2012.
molecular biology. The New Microbiologica, 2017.
[PE12] Alexander Ploss and Matthew J. Evans. Hepatitis c virus host cell
[KA14] Jennifer A. Kanakry and Richard F. Ambinder. Ebv-related lym- entry. Current Opinion in Virology, 2012.
phomas: New approaches to treatment. HHS Author Manuscripts,
2014. [PGS+ 15] Maria Raffaella Petrara, Silvia Giunco, Diego Serraino, Riccardo
Dolcetti, and Anita De Rossi. Post-transplant lymphoproliferative
[KAS17] Troy Kish, Andrew Aziz, and Monica Sorio. Hepatitis c in a new disorders: from epidemiology to pathogenesis-driven treatment.
era: A review of current therapies. P and T : a Peer-Reviewed Cancer Letters, 2015.
Journal for Formulary Management, 2017.
[PSR+ 15] Aryn A. Price, Timothy R. Sampson, Hannah K. Ratner, Arash
+
[KKG 14] Edward M. Kennedy, Anand V. R. Kornepati, Michael Goldstein, Grakoui, and David S. Weiss. Cas9-mediated targeting of viral
Hal P. Bogerd, Brigid C. Poling, Adam W. Whisnant, Michael B. rna in eukaryotic cells. Proceedings of the National Academy of
Kastan, and Bryan R. Cullen. Inactivation of the human papillo- Sciences of the United States of America, 2015.
mavirus e6 or e7 gene in cervical carcinoma cells by using a bacte-
rial crispr/cas rna-guided endonuclease. Journal of Virology, 2014. [SEKI+ 16] Caecilia H.C. Sukowati, Korri E. El-Khobar, Susan I. Ie, Beatrice
Anfuso, David H. Muljono, and Claudio Tiribelli. Significance of
[LBZ+ 17] Cody S. Lee, Elliot S. Bishop, Ruyi Zhang, Xinyi Yu, Evan M. hepatitis virus infection in the oncogenic initiation of hepatocellular
Farina, Shujuan Yan, Chen Zhao, Zongyue Zeng, Yi Shu, Xingye carcinoma. World Journal of Gastroenterology, 2016.
Wu, Jiayan Lei, Yasha Li, Wenwen Zhang, Chao Yang, Ke Wu,
Ying Wu, Sherwin Ho, Aravind Athiviraham, Michael J.Lee, [SKJL10] Pamela L. Shaw, Austin N. Kirschner, Theodore S. Jardetzky, and
Jennifer Moriatis Wolf, Russell R. Reid, and Tong-Chuan He. Richard Longnecker. Characteristics of epstein-barr virus envelope
Adenovirus-mediated gene delivery: Potential applications for gene protein gp42. Virus Genes, 2010.
and cell-based therapies in the new era of personalized medicine.
Genes and Diseases, 2017.
71 72
[SOC+ 17] C. Signorelli, A. Odone, V. Ciorba, P. Cella, R. A. Audisio, A. Lom-
bardi, L. Mariani, F. S. Mennini, S. Pecorelli, G. Rezza, G. V. Zuc-
cotti, and A. Peracino. Human papillomavirus 9-valent vaccine for
cancer prevention: a systematic review of the available evidence.
Epidemiology and Infection, 2017.
Predicting NBA Playoffs Using Machine Learning
[WHK15] Martin K. White, Wenhui Hu, and Kamel Khalili. The crispr/cas9 ∗
Sean Liu
genome editing methodology as a weapon against human viruses.
Discovery Medicine, 2015. April 2, 2021
[WQ14] Jianbin Wang and Stephen R. Quake. Rna-guided endonuclease
provides a therapeutic strategy to cure latent herpesviridae infec-
tion. Proceedings of the National Academy of Sciences of the United Abstract
States of America, 2014.
This project attempts to predict the NBA playoff bracket using ma-
[YAM07] Lawrence S. Young, John R. Arrand, and Paul G. Murray. Chap- chine learning methods. It will consider one self-constructed model and
ter 27 ebv gene expression and regulation. Human Herpesviruses: one machine learning model built from various machine learning algo-
Biology, Therapy, and Immunoprophylaxis, 2007. rithms. The project will also determine the most efficient model for pre-
dicting NBA results and which way to select data gives an accurate and
[ZB06] Zhi-Ming Zheng and Carl C. Baker. Papillomavirus genome struc- consistent prediction. Finally, the project will investigate the effect of
ture, expression, and post-transcriptional regulation. HHS Author home and away variables on the teams’ performance and the model’s ac-
Manuscript, 2006. curacy.
[ZKY+ 12] Baochun Zhang, Sven Kracker, Tomoharu Yasuda, Stefano Ca-
sola, Matthew Vanneman, Cornelia Hömig, Hölzel, Zhe Wang, 1 Introduction
Emmanuel Derudder, Shuang Li, Tirtha Chakraborty, Shane E.
Cotter, Shohei Koyama, Treeve Currie, Gordon J. Freeman, Jef- The National Basketball Association (NBA) is considered as the premier bas-
fery L. Kutok, Scott J. Rodig, Glenn Dranoff, and Klaus Rajewsky. ketball league for professional male basketball players in USA. It is made up of
Immune surveillance and therapy of lymphomas driven by epstein- 30 teams, split into the Eastern and Western conferences [Aut01].
barr virus protein lmp1 in a mouse model. Cell, 2012. During the playoff, the top 8 teams from each conference (Eastern and West-
ern) are chosen to compete for the championship. The rankings are decided
based on the teams’ performances during the regular season. Then, the teams
play against each other with the 1st place playing against the 8th place, the
2nd place playing against the 7th place, etc. Each game will be a best-of-seven
match, and teams will rotate between home and away.
Like AlphaGo in the Go Contest [Sil02], machine learning is a well-known
prediction tool for complex process. The question of this research is, can ma-
chine learning be used to predict the NBA playoff bracket? And what is accu-
racy of such prediction compared with the real results? What machine learning
method is the best solution for NBA playoff bracket prediction?
To better analyzing and comparing the performance of the machine learning
in predicting the NBA playoff bracket, firstly we create a self-made prediction
model based on several key game variables that will impact the game result
mostly by our best knowledge about the NBA games. These variables include 1)
effective field goal percentage, 2) free throw percentage, 3) turn over percentage,
4) Offensive rebound percentage, 5) Defensive rebound percentage. And We
∗ Advised by: Derek Sorensen, University of Cambridge
73 74
focused on working out the probability of Team A winning against Team B,
then applying this to every game in the playoff. Base on the historic game data
of 2014 to 2018, the playoff prediction of 2018,2017 and 2016 are carried out
and comparison with the real playoff brackets are also presented.
As for the machine learning model for the NBA playoff bracket prediction,
here, we are focusing on 5 different machine learning models that have already
been implemented in the Python Machine Learning Library (Scikit-learn), i.e.,
Logistic Regression (LR), Linear Discriminate Analysis (LDA), Support Vector
Machine (SVM), K-Nearest Neighbors (KNN) and Classification and Regression
Tree (CART) [Dhi03] - [TA11]. For comparison with the self-made model, the
same playoff prediction of 2018, 2017 and 2016 are carried out and comparison
among different machine learning models are given accordingly.
2 Result
2.1 Exposition of self-made prediction model
Based on the testing results for our self-made prediction model, we have the Figure 2: 2018 NBA Playoff (Original)
following prediction results (Table 1). And the predicted playoff bracket with
the original ones are shown in Figure 1 and 2, with the prediction difference
highlighted in red color.
Figure 3: Playoff Prediction Accuracy of Self-made Prediction Model
prediction, we used teams’ statistics over 4 years from 2014-2018. For the 2017
playoff prediction, we used statistics of teams over a time of 3 years from 2014
- 2017. Finally, for the 2016 playoff prediction, we only used statistics of teams
over two years from 2014-2016.
2.2 Exposition of the Machine Learning Models

In order to evaluate the performance of machine learning in NBA playoff bracket
prediction, 5 different machine learning models are employed, which have been
implemented by Python Machine Learning Library (Scikit-learn), i.e., 1) Logis-
tic Regression (LR), 2) Linear Discriminate Analysis (LDA), 3) Support Vector
Machine (SVM), 4) K-Nearest Neighbors (KNN) and 5) Classification and Re-
gression Tree (CART) [Dhi03] - [TA11]. With two different training data selec-
tion methods and home/away investigation, the most accurate machine learning
Figure 1: 2018 NBA Playoff (Prediction)
model for NBA playoff bracket prediction is finally presented.
There are two ways in which we chose to select the data to train the al-
In summary, the self-built prediction model performed best when predicting
gorithm. 1. The first method was to consider each Team’s performance when
the 2018 playoff, getting an accuracy of 80%. The second-best prediction was
playing against all other teams 2. The second method was to evaluate a team’s
the 2017 prediction, obtaining an accuracy of 66.7%. The worst prediction was
performance with one other s pecific Team t o determine i ts win r ate against t hat
for t he 2016 bracket, only getting an accuracy of 53.3%. For t he 2018 playoff
75 76
specific Team In summary, selecting only the data where two teams played against each
other resulted in inaccurate and inconsistent predictions. It also means that
2.2.1 Method 1 – Selecting all data the models’ accuracy in this data selection method will not be considered when
calculating the best performing model due to the inconsistency and inaccuracy.
In this model, we trained the different machine learning algorithms with all the Therefore, it can be concluded that selecting all data is a better data selection
statistics of every Team. Like the self-constructed model, the 2018 prediction method.
used data over four years, the 2017 prediction used data over three years, and
the 2016 prediction used data only over two years.
From the above data, algorithms generally performed relatively well and
consistent in 2018 and 2017, except for the SVM model and the LR model
(Figure 3). The SVM model generally had a low and consistent prediction
accuracy in the three years, and the LR did significantly better in 2017 than in
2018.
Overall, LDA had the highest mean accuracy of 71.2%, followed by the
CART model, with a mean accuracy of 69.2%. The worst performing model is
the SVM algorithm with an accuracy of 0.436 only.
Figure 5: Playoff Prediction Accuracy of Different Machine Learning Models

with partial data selected and average 30 trial runs
2.2.3 Home and Away Investigation

The home and away variables are widely considered an essential variable on a
team’s performance and players. The home-court will have more fans, and the
positive atmosphere will give the home team a spiritual boost, which may result
in a better performance.
Figure 4: Playoff Prediction Accuracy of Different Machine Learning Models An experiment is conducted by training and running the program three times
with all data selected and average 30 trial runs with different data. One will have all data from home and away games, another
will have only the data from home games, and the final one will have only the
data from away games. Differences between the predicted results are analyzed
2.2.2 Method 2 – Selecting Partial Data and evaluated.
Based on the above data (Table 2), generally, teams had much better per-
In this model, we trained the machine learning models with a partial amount formance and a higher win percentage when playing as the home team.
of data, which is only based on one Team’s performance against a specific op- To summarize, the home and away variable greatly influenced teams’ per-
ponent. In other words, it is the data between two teams that we are trying to formance level and win percentage in general. In theory, the variable should not
predict. affect model accuracy to a significant extent. But in this case, the model did
Based on the data collected (Figure 4), there are no clear trends or patterns impact the model accuracy, which can be caused by other factors in real life.
available. All the model predictions have significantly large variations in each
year. It also doesn’t have a clear correlation to data size. Although 2016 was
again the worst year of prediction, the 2017 prediction did significantly better
than the 2018 prediction, proving that there are no trends.
77 78
3.2 Machine learning prediction model
Similar as the results for the self-made prediction model, the prediction accuracy
of 2018 is the highest while the one of 2016 is lowest when all data are selected
to train the machine learning models (Figure 3). All algorithms except the SVM
model performed significantly worse in 2016. One hypothesis is that the models
reached a tipping point in 2016 when the data size is not big enough to support
accurate predictions.
Figure 6: 2018 LR Prediction Results
2.2.4 Most Accurate Machine Learning Model for NBA Playoff Pre-
diction
To conclude, the most accurate machine learning model at predicting the NBA
playoffs is LDA, which reached an accuracy of 71.2%. The performance of
the models at predicting with a partial amount of data is neglected since it is
considered that the data selection did not give useful information.
3 Discussion
3.1 Self-made prediction model
Based on the model’s accuracy and the size of the data, we see a trend between
the two variables, with 2018 having the largest dataset and the highest model
accuracy and 2016 having the smallest dataset and the lowest model accuracy
Figure 7: 2016 Team Performance Prediction with All Data Selected
(Table 1). One possible explanation for the model’s changing performance is
that it works well with larger datasets while having lower performances when
Another hypothesis is that there is an error in the program itself that is
working with smaller datasets.
causing the 2016 prediction to deviate. This can be seen through the models’
Another possible reason is that the model is simply not consistent in predic-
accuracy in 2016 (Table 3), except for SVM, all had an accuracy of 53.3%.
tion. It might be a coincidence that there is a correlation between data size and
Here, it demonstrated that each model except for SVM had the same playoff
model accuracy since we only have data for three years of prediction. Further
prediction for every round. Although teams have slightly shifting percentages
investigation can be carried out to confirm the effect of the data size on the
in different models, which may symbolize that there isn’t an error and that all
model’s accuracy. This can be done by running the prediction model for more
models are independent of each other, it is still doubtful that each model had
years with different data sizes to understand the correlation between the two
the same win percentage and same prediction. This will be a research question
variables better.
for future investigations to confirm if there is an error in the program causing
the deviation in 2016, or the model reached a tipping point in data size that is
causing the variation to occur.
79 80
For model training with partial data selection (Figure 4), it can be concluded
that partial data selection method gives inaccurate and inconsistent predictions.
This is likely because there is very little data for a given pair of teams. To be
specific, two teams only play against each other ten times a year, with Team1
playing as the home team for five matches and Team2 playing as the home team
for another five games. Additionally, only 80% of the data are used to train,
meaning that only eight sets of data are provided for training each year. This
resulted in inaccurate predictions with inadequate dataset. It also means that
the prediction models will be more likely to give the two teams a 50 percent
win rate each due to the small amount of data for testing and training. This
Figure 8: NBA historic statistics dataset format and headers
will result in the program randomly selecting a winner between the two teams,
making the prediction model inconsistent.
Additionally, the two sets of predictions namely home and away should have 4.1 Self-made prediction model
similar accuracy theoretically. This is because teams typically have a higher To start, we first created our own prediction model to predict the NBA bracket.
win percentage when playing as the home team and a lower win percentage We focused on working out the probability of Team A winning against Team B,
when playing as the away team. If all teams perform better when playing as the then applying this to every game in the playoff.
home team, they should get roughly the same increase in performance level, so We have to narrow our focus on specific game variables, which significantly
it should not affect the accuracy to a significant extent. This is the same when impact the game result. After some research, we decided to use the following
teams are playing as the away team. They should all perform relatively worse, variables:
so the models’ accuracy should not shift by a significant amount. 1)EFG% effective field goal percentage [Aut12], considers both 2pts field
However, in this case (Table 2), the model accuracy did shift significantly, at goals and 3pts field goals in one variable and considered their weight with three-
13%. This is due to outliers like the team MIN, which had a better performance pointers worth 1.5 times of a two-pointer.
when playing as the away Team than as the home team. It is also because 2)FT% free throw percentage [Aut12], calculates the percentage of free-throw
different teams had different performance levels when playing as the home team. makes for a specific team.
For example, GSW had an increase in a win percentage of 30% when playing as 3)TOV% turn over percentage [Aut12], is an estimate of turnovers by a team
the home team. On the other hand, team PHI only had a 9% increase in win per 100 possessions.
percentage when playing as the home team. One hypothesis is that GSW has 4)ORB% Offensive rebound percentage [Aut12], is an estimate of the per-
more fans than other teams, so they have a better atmosphere when playing as
centage of offensive rebound that a team gets.
the home team. However, many other factors can decide a team’s performance
5)DRB% Defensive rebound percentage [Aut12], is an estimate of the per-
when playing as the home team and the away team. These factors can be further
centage of defensive rebounds taken by a team.
investigated in the future.
(2 point f ield goals made + 1.5 ∗ 3 point f ield goals made) ∗ 100
EF G% =
4 Methods T otal f ield goals made
To test the effectiveness of our self-made NBA playoff prediction model and all f ree throws made ∗ 100
FT% =
the related machine learning algorithms, certain NBA historic statistics data f ree throws attempted
from 2014 - 2018 are needed, which can be access from many open source NBA
statistics. And these historic NBA statistics are usually saved as .csv file format, number of turnovers ∗ 100
T OV % =
which we can use the Python pandas library read-csv module to load the dataset f ield goal attempted + 0.44 ∗ f ree throws attempted + number of turnovers
from the corresponding csv URL link, the format and header of the dataset is
of the following form (in Figure 5.) of f ensive rebounds ∗ 100
ORB% =
of f ensive rebounds + opponent def ensive rebounds
def ensive rebounds ∗ 100

DRB% =
def ensive rebounds + opponent of f ensive rebounds
81 82
4.1.1 Algorithms 6 – The last case is when the two sets of data have the same upper limit
(yB=yA). Then Pi = 0.5 in this case.
The five variables that were chosen are considered the most impactful factors
in the game. The second step of our model is to decide on the algorithm we are
going to use to calculate the probability of Team A beating Team B; the chosen 4.2 Predicting the playoff bracket
algorithm was: In order to predict the playoff bracket, we created Python’s function to calculate
the probability of Team A defeating Team B, and we applied it to predict the
Pwin = c1 P1 + c2 P2 + c3 P3 + c4 P4 + c5 P5 playoff bracket for 2018.
Here, ci is the proportional correlation of the variable vi with winning. In
other words, the larger the value of ci , the more variable vi will contribute to
the winning of a game. Pi is the probability that Team A will have a higher
score than Team B on variable vi . By multiplying the probability of the two
factors together and adding all the numbers up for all five different variables,
we predict Team A beating Team B in a match.
4.1.2 ci calculation
The formula for ci is:
ri
ci =
r1 + r 2 + r3 + r4 + r 5
Here, ri represents the Pearson correlation coefficient of the variable vi with
winning. However, winning is a categorical value that cannot be used in the
Pearson correlation. Therefore we decided to represent winning with the point
difference between the two teams.
4.1.3 Pi calculation
To calculate the value for Pi , we used the principle of confidence intervals, which
is defined to be the probability that a parameter will fall between two sets of
values with a specific confidence level. [Wil13]
1 - Calculate a 95% confidence interval of for both teams Figure 9: Python Code Snippet for Self-made Prediction Model
2 - We defined the confidence intervals for Team A as [xA ,yA ] and Team B
as [xB ,yB ] In the python code snippet (Figure 6), the function select team(), which
3 – The first case is when the intervals don’t overlap. In this situation, the predicts the winner between Team A and Team B, is called many times. This
Team with the higher interval has a 95% chance of scoring higher. (Note: The calculates the winners for the quarter-final, the semi-final, the finals, and in
percentage might be slightly higher than 95%, but in this case, we consider it the end, it calculates the winner of the year. This predicted playoff is then
as 95%.) appended to a list and compared to the actual result of the 2018 playoff to cal-
4 - The second case is when the two intervals overlap, and Team A has a culate a prediction accuracy. The original playoff is pre-loaded into the program
higher upper limit (yA ¿ yB ). Here, the formula to calculate Pi is: beforehand.
yA − y B 4.3 Machine learning prediction models

Pi = 0.95
yA − x B
In order to test if machine learning algorithm can be used to predict the NBA
5 - The third case is when Team B has a higher upper limit (yB ¿yA ). Here, playoff bracket and evaluate which machine learning model has the best predic-
the formula to calculate Pi is: tion accuracy, 5 different machine learning models that have been implemented
yB − yA in Python Scikit-learn machine learning library are employed with two linear
Pi = 1 − 0.95 (LR and LDA) and three nonlinear (KNN, CART and SVM) ones (Figure 6).
yA − x B
83 84
After loading the dataset from the historic NBA statistics CSV file, depend-
ing on the two different data selection methods, all data or partial data, together
with home or away analysis, data related to year of 2016, 2017 and 2018 for NBA
playoff prediction can be split into different data arrays, so that the prediction
accuracy of different machine learning models can be analyzed accordingly.
To test the prediction accuracy for each different machine learning model,
the dataset needs to be split into two sets, one for the model training and one
for the model prediction on 8:2 randomly selection basis, which means 80% of
the data will be used as training data and 20% will be used to evaluate the
prediction accuracy and the data is randomly selected.
After the dataset is split for training and validation, the fit function for each
machine learning model will be called to train each individual models. After
model training, the predict function for each machine learning model will be
called to make the final prediction based on the validation dataset generated
before and the prediction accuracy for each models will also be calculated by
the accuracys core function (Figure 7).
Note that in Python, loc function is a frequently used function to retrieve
partial data in the dataset related to certain variable value, like certain year, Figure 11: Python code snippet for dataset split and model cross-evaluation
certain team, etc.
References
[Anu06] Mehta Anukrati. A beginner’s guide to classification and regression
trees. https://www.digitalvidya.com/blog/classification-and-regression-
trees/, 0006.
[Aut01] No Author. National basketball association.
https://en.wikipedia.org/wiki/NationalBasketballAssociation, 0001.
[Aut09] No Author. Machine learning - logistic regression.

https://www.tutorialspoint.com/machinelearningwithpython/machine-
learningwithpythonclassificationalgorithmslogisticregression.htm, 0009.
[Aut10] No Author. Advantages and disadvantages of logistic re-
gression. https://iq.opengenus.org/advantages-and-disadvantages-of-
logistic-regression/, 0010.
Figure 10: Python code snippet for importing modules, functions and models
[Aut12] No Author. Glossary. Basketball Reference, No Date,
https://www.basketball-reference.com/about/glossary.html#tovpct,
0012.
5 Acknowledgement
[Cor05] Maklin Cory. Linear discriminant analysis in python.
I would like to thank Mr. Derek Sorensen of Horizon Academic Research Pro- https://towardsdatascience.com/linear-discriminant-analysis-in-
gram (HARP), together with my parents, my sister for the support. python-76b8b17817c2, 0005.
[Dhi03] K Dhiraj. Top 5 advantages and disadvantages of decision

tree algorithm. https://medium.com/dhiraj8899/top-5-advantages-
and-disadvantages-of-decision-tree-algorithm-428ebd199d9a, 0003.
85 86
[Jos08] Starmer Josh. Linear discriminant analysis (lda) clearly explained.
https://www.youtube.com/watch?v=azXCzI57Yfc&t=516s, 0008.
[Nar04] Kumar Naresh. Advantages and disadvan-

tages of knn algorithm in machine learning. Extending the Drift Diffusion Model to the
http://theprofessionalspoint.blogspot.com/2019/02/advantages-and-
disadvantages-of-knn.html, 0004. Cognitive Realm
[Rus07] Pupale Rushikesh. Support vector machine (svm) – an overview. Hriday Nayak ∗
https://towardsdatascience.com/https-medium-com-pupalerushikesh-
svm-f4b42800e989, 0007. April 2, 2021
[Sil02] David Silver. Mastering the game of go without human knowledge.
ARTICLE, doi:10.1038/nature24270, 0002.
[TA11] Ibrahem Abdlehameed Hassanien Aboul Ella Tharwatt Alaa, Abstract

Gaber Tarek. Linear discriminant analysis: A detailed tu-
The drift diffusion model has had great success as a
torial. https://www.researchgate.net/publication/316994943Linear-
discriminantanalysisAdetailedtutorial, 0011. computational model to study the underlying processes of
sensory and value-based decision making and the diffu-
[Wil13] Kenton Wil. Confidence interval. sion process may actually mimic how the brain integrates
https://www.investopedia.com/terms/c/confidenceinterval.asp, 0013.
evidence and makes decisions. Here I review recent ap-
plications and extensions of the drift diffusion model to
self-control, loss aversion, driving behaviour, racial biases
and reinforcement learning with the aim of finding out
whether the model is applicable to more cognitive tasks
1 Introduction
Decisions are ubiquitous. Every day, we make thousands of decisions, ranging
from automatic low-level tasks like whether or not to look at a section of the
screen to high-level tasks which require more deliberation like which movie to
watch. In the last thirty years, major progress has been made in understanding
how decisions are made in the brain.To explain any aspect of decision mak-
ing, experiments are designed and then computational models are built on this
experimental data. To make the models more cohesive and increase their ex-
planatory power, brain data is included to see how the deliberation corresponds
to activity in the brain. Out of a plethora of methods, electrophysiology1 and
∗ Advised by: Julian Day-Cooney, The University of Chicago, Oregon Health and Science
University
1 Extracellular Electrophysiology - A recording technique which involves the insertion of
electrodes into the brain. It measures the change in the electrical activity in the neurons near
the electrode and thus measures firing rate of neurons in spikes per second. It gives a great
temporal resolution making it a very powerful technique. However, it is an invasive method
and can only be used to measure the activity of a few neurons at a time.
87
88
functional magnetic resonance imaging (fMRI)2 are specifically used to mea-
sure firing rates and to gauge where the decision-related brain activity is taking
place, respectively.
Many real-life decisions require accumulation of evidence and information,
either from the environment or from our memories, until it passes a threshold.
This accumulation-to-threshold is explained by a group of models called sequen-
tial sampling models. Through extensive research, one of these models stands
out as the most effective: The Drift Diffusion Model (DDM).
The DDM (Ratcliff,1978) [Rat78] postulates that decisions are made by the
accumulation of noisy evidence over time which terminates once it reaches a
threshold or a bound. The decision threshold is the amount of evidence needed
to choose an alternative and make a decision. In the DDM, there is only one
accumulation process whereas in other accumulator models the evidence for
each response is accumulated independently. These models are like a race.
The accumulation process that reaches the threshold first is what the subject
decides. In the DDM, evidence accumulation is competitive. Figure 1 shows the
drift diffusion process for a perceptual discrimination task as well as an ideal
accumulator model (also called a race model).
The beauty of the DDM is that the brain makes decisions as it has been
portrayed in Figure 1. The DDM is not just a model made to explain decision
making but could be how the brain integrates evidence and makes decisions.
The simple DDM is defined by four parameters, the starting point, the drift
rate(µ) i.e. the rate of evidence accumulation, the value of the bounds (a and
b), and the non-decision time, also called latency time which is the sum of the
time before initiation of the accumulation and the time taken for action once a
decision is reached. (Ratcliff,1978) [Rat78]
This review will begin with an in-depth description of the DDM, its advan- Figure 1: A) z is the starting point for the process, a and -b are the thresh-
tages and the research on the neuronal populations representing the subparts of olds. In the figure, there are two wobbly lines, which represent the decision
the DDM. Following this, I will discuss the applications of the DDM to different process. The decision is encoded in a decision variable and this variable
domains, with the aim of looking at its performance in more cognitive tasks. ‘drifts’ towards a threshold. A decision variable is a quantity which defines
The review will conclude with possible future avenues. the possibility of one alternative over another and is driven by the integra-
tion of evidence over time. It can be thought of as a link between sensory
evidence and the final choice. The straight lines connecting the starting
1.1 Why DDM? point z and the thresholds are the drift rates. The image at the top is
Certain aspects of the Drift Diffusion Model have made it very successful in all represented by the blue lines while the image at the bottom is represented
its applications so far. This section looks at the advantages of using the DDM by the red lines. In the figure, the blue lines have a much larger drift rate
as an analytic tool. The DDM explains response times (RTs) very well for and as a result a short response time. Although intuitively we know that
easier decisions will be made faster, this figure and the DDM, in general,
both correct and error responses. It helps us visualize the effects of attributes
gives us computational proof on the relation between decision difficulty and
like task difficulty and time pressure on the RTs, which serve as an important response time. Figure adapted from Heekeren et al., (2008) [HMU08]B)
tool in analyzing behaviour. This explanatory power helps to differentiate the The graph adjacent to the race model shows two accumulation processes,
2 Functional Magnetic Resonance Imaging (fMRI) - An imaging technique that measures one for each response for the race models, whereas, in the DDM, there is
brain activity by detecting changes in blood flow. It is used to measure BOLD signals (blood- only one accumulation process which is a competition between the two al-
oxygen-level-dependent signals). It gives us great spatial resolution and can provide a clear ternatives. Figure adapted f rom Summerfield and Koechlin,(2008) [ SK08]
image of how brain activity is localised. Another advantage is that it is non-invasive and is
a relatively safe technique. However, it gives poor temporal resolution and does not show us
moment to moment changes in activity.
89 90
DDM from other sampling models like random walk models and accumulators
which cannot model the RTs as accurately as the DDM (Ratcliff, 2004 [Rat04]).
Although these conclusions are intuitive (harder tasks will have longer RTs), the
DDM provides a computational framework for these conclusions. The response
times are captured by RT distributions. These distributions can be represented
by curves above their respective thresholds. The distribution encloses all the
possible RT values for a particular experiment and its shape shows the variability
in the response times and thus the variability in drift rates. The curves are
shifted or skewed when task difficulty is changed or a time pressure is applied.
Thus, the distributions help in giving an insight into the change in performances
when task attributes are altered. Another advantage the DDM provides is that
it explains the speed-accuracy trade-off well, (for an in-depth review see Bogacz
et al., 2010 [BWFN10]). Higher decision thresholds will lead to more accurate
answers since they require more evidence but will also lead to greater response
times. On the other hand, lower thresholds will lead to fast responses, however,
will result in a greater error rate. (see Figure 2A). Consider an investigation -
more solid evidence will lead to catching the correct perpetrator but will take
more time. However, quick justice could result in catching the wrong person.
The DDM also provides a better understanding of choice biases. Biases can
occur in two ways. First, there are starting point biases i.e. the starting point
is closing to one bound, thus the decision-maker is inherently biased towards
one decision. Second is the drift rate bias, in which the drift rate is higher for
one response, biasing the decision to that alternative. Figures 2B and 2C show
the effect of both these biases on the diffusion process.
Figure 2: a)The relationship between the speed-accuracy trade-off and
the decision threshold. Lower thresholds can result in less accurate
2 Neural Correlates decisions. b) This figure shows the effects of the starting choice bias
on the diffusion process. c)The effect of the drift rate bias on the dif-
While understanding the working of the brain is important, a major goal of fusion process (Figures adapted from Mulder et al., 2014 [MvMF14])
neuroscience has been to map these processes to underlying circuits in particular
regions of the brain. These regions are called neural correlates. This section will
look at various studies pinpointing the correlates for the different sub-process the drift rate was significantly reduced under the influence of the TMS while the
of the diffusion process: evidence accumulation (drift rate), decision threshold, non-decision time was almost unaffected, thus, showing the role of the DLPFC
starting point bias, and comparison of alternatives. To find neural correlates in in evidence accumulation. Other studies have reported that areas like the frontal
humans, the tool of choice would be fMRI, since it is non-invasive. However, eye field (FEF) and intraparietal sulcus (IPS) could be responsible for evidence
causal studies3 are also performed, giving more definitive proof that an area is accumulation. (Basten et al., 2010 [BBHF10]; Ho et al., 2009 [HBS09]; Liu and
responsible for a sub-process. Using fMRI, Rolls et al., (2010) [RGD10] found Pleskac, 2011 [LP11])
signatures in the dorsolateral prefrontal cortex (DLPFC) which could represent The lateral intraparietal area (LIP), a subdivision of the IPS, has been also
evidence accumulation. Philiastides et al., (2011) [PAHB11] showed the causal shown to represent sensory integration (Roitman and Shadlen, 2002 [RS02]).
role of the DLPFC using trans-cranial magnetic stimulation.4 They found that At this time, research points to a frontoparietal network (a network of areas
3 Causal Studies - Causal techniques are used to find direct causal relationships between in the frontal and parietal lobes of the brain) that is responsible for evidence
brain regions and a specific function. Causal methods include inhibition of a particular area accumulation. Studies regarding the decision threshold have pointed to a fron-
by stimulation, pharmacological inactivation and lesion studies. They have great explanatory tostriatal network which would include the anterior cingulate cortex (ACC),
power in finding neural correlates. striatum, and the pre-supplementary motor area (pre-SMA) as candidate areas
4 TranscranialMagnetic Stimulation (TMS) - It is a non-invasive procedure in which neu- (Forstmann et al., 2008 [FDB+ 08]; Ivanoff et al., 2008 [IBM08]; Van Veen et
rons in the brain are stimulated by a magnetic field which induces electrical activity in those al., 2008 [VVKC08]; Winkel et al., 2012 [WvMR+12]).
neurons.
92
91
Kiani et al., (2014) [KCRN14] showed the response of neurons in the pre-
arcuate gyrus during Changes of Mind. A change of mind would be a sudden
change in the direction of the evidence accumulation. If the decision variable
is drifting towards the upper threshold, a change of mind can be seen in a
sudden reversal of direction towards the lower threshold. Mathematically, the
sign of direction changes. The firing rates of these neurons peaked just before
the saccade which could indicate the encoding of the decision threshold.
Various studies looking at value-based decision making have found encoding
of subjective value and choice bias in the orbitofrontal cortex (OFC) (Forstmann
et al., 2010 [FBD+ 10]; Padoa-Schioppa and Assad,2006 [PSA06]; Summerfield
and Koechlin, 2008 [SK08]). Other frontal areas such as the ACC, ventromedial
prefrontal cortex (VMPFC), and DLPFC have been shown to encode starting
point bias (Mulder et al., 2012 [MWR+ 12]). The above-mentioned areas have
also been shown to be responsible for the comparison of alternatives in choice
tasks (Hare et al., 2011 [HSC+ 11]; Hunt et al., 2012 [HKS+ 12]).
Figure 3 summarizes the current research in finding neural correlates. Each
dot in the figure represents a group of studies. The size of the dot represents
the number of studies conducted. Thus, the figure shows every study conducted
for the different parameters and sub-processes. The location of the dot shows
the areas that are responsible for a sub-process. The colour of the dot shows Figure 3: Each dot is a separate group of studies. The size of the dot
the region of the brain the specific correlate is situated in and each region is gives the number of studies conducted in a particular region. Re-
represented by a unique colour as shown in the legend. gions have been highlighted as given in the legend. studies have been
conducted to find correlates for evidence accumulation. This figure
3 DDM In Perceptual and Lexical Tasks shows a frontoparietal network for the accumulation and a frontostri-
atal network for decision threshold. The starting point bias is almost
This section will delve into the two most successful domains of applications only encoded by frontal networks. However, areas can have multi-
of the DDM to behavioural data: perceptual tasks and lexical tasks. In the ple functions and the distinctions are not always concrete. (Figure
realm of perceptual decisions, researchers have applied the DDM to a simple adapted from Mulder et al., 2014) [MvMF14]
dot motion-discrimination task (also called Newsome Dots or the Random Dot
Kinematogram) and a categorization task. In the RDK task, the subject is
sion task. In this task, the human subject has to categorize the given stimuli
shown a group of moving dots and needs to choose the average direction of the
into words and non-words. The stimuli were variable and were taken from a set
dots by a saccade. This task gives great control over task difficulty. It introduces
of high-frequency words, low-frequency words, very low-frequency words, pseu-
motion perception and also requires the subject to compute the average motion
dowords, and non-words. The model explained the RTs for, both correct and
of the dots which requires a large amount of integration. Figure 4 shows an
error responses, and the probability of getting the decision correct very well for
RDK task.
all types of stimuli. (see Table 3 in Ratcliff et al., 2004 [Rat04])
Figure 1 shows an example categorization task, where the subject was re-
Recently, these applications have been extended to look at aging and IQ
quired to place the given image in the house or the face category. This task
from a unique perspective. Studies have shown that older adults are slower
also gives great control over changing the difficulty of the task but also requires
than young adults due to longer non-decision times and a higher boundary,
evidence accumulation over time, and on the difficult trials, requires the com-
although age does not play a role in drift rate. This makes intuitive sense
parison of the alternatives shown to the ideal response. For example, if the
since older adults are usually more cautious and the decision thresholds prove
image has a low contrast the subject would need to compare the image to an
this (Theisen et al., 2020 [TLvKV20]). Studies can also show how differences
ideal image of a face and that of a house and try to give the correct response.
in IQ can affect decisions. Subjects with a higher IQ have higher drift rates
Both these tasks have been successfully modelled by the DDM. (see Gold and
but have almost equal non-decision times and boundary separations, as normal
Shadlen, 2007 [GS07]; Heekeren et al., 2008 [HMU08] for an extensive review)
subjects. ( Ratcliff et al., 2010 [ RTM10]; Ratcliff and Mckoon, 2011 [ RM11]).
Ratcliff et al., (2004) [RMG04], showed the DDM applied to a lexical deci-
93 94
quantitative description of their data and could be the model used by the brain.
They conducted this analysis on a subjective value task in which the subjects
had to choose between two food options.
However, the fDDM, as described in the paper, was suitable only for binary
choice and did not take attention or fixations into account. When we choose be-
tween many alternatives, we often foveate (position our fovea centralis, the part
of the eye with the sharpest vision) on the preferred option and this can bias
our choice. This was not incorporated in the fDDM. To change this, Krajbich
and Rangel, (2011) [KR11], proposed a novel drift-diffusion process for subjec-
tive multi-alternative decisions, the Attentional Drift Diffusion Model (aDDM).
Their model included a fixation bias and explained the data for both binary and
trinary value-based choices. The aDDM can also be extended to simple pur-
Figure 4: An RDK task for a macaque monkey. The coherence of
chasing tasks, in which subjects need to decide whether or not to buy a product
motion can be changed trial to trial. The monkey has to gauge the
for the given price (Krajbich et al., 2012 [KLCR12]). The model explained RTs
average motion and indicate its response by a saccade to one of the
for different sets of stimuli but also showed the adverse impact of visual fix-
two targets on the screen. (Figure adapted from Heekeren et al.,
ations. When subjects looked at the product more, they were more likely to
2008) [HMU08]
buy it. On the other hand, if they looked at the price for a longer period, they
were more likely to reject it. The effects of visual fixations and affection thus
Studies looking at sleep deprivation, clinical populations, alcohol consumption, seem to bridge the perceptual and economic domains together. Thus, we can
and reduced blood sugar have had success using a diffusion model analysis, thus see that the DDM has been successfully modified to purchasing decisions and
proving that the DDM can be clinically useful. (See Forstmann et al., 2016 value-based choices, both for binary and trinary choices. Attempts have been
[FRW16] for an excellent review). made to extend the aDDM to quaternary choice (von Boguslawski and Mildén,
2015 [vBM15]), with mixed results. Although they have modeled choice well,
the sample size may be small and the results may not be very significant. Still,
4 Extending the DDM to Economic Choices it is another step towards modeling more complex tasks.
This section will look at the modifications of the DDM for it to be applied to
subjective tasks. All the tasks mentioned in the paper so far have had a defined 5 Cognitive Tasks
correct response. This section will be an introduction into the domain of value-
based choice. As I stated before, the model looked at so far is the This section will look at novel applications of the DDM to more cognitive tasks.
simple DDM (sDDM) with 4 parameters. To extend the DDM to economic and The purchasing and value-based experiments talked about in previous experi-
subjective choices its computational framework behind the model needs to be ments were simplified accounts of real-life decisions. This section will address
modified. Milosavljevic et al., (2010) [MMH+10], compared the sDDM with 3 of experiments looking at more complex behaviors and decisions.
its variants – the simple collapsing barrier DDM (scbDDM) 5, the full DDM
(fDDM) 6, and the full collapsing barrier DDM (fcbDDM) 7, using the 5.1 Self-Control
Bayes Information Criterion 8. They found that the fDDM provided the best
Berkman et al. (2017) [BHL+ 17], put forth an alternative model for self-
5SimpleCollapsing Barrier DDM (scbDDM) - It is a modification of the simple DDM in control. Rather than a competition between the impulsive and deliberative
which the bound values (a and b) decrease as time progresses thus reducing the amount of
evidence needed in the later stages of the trial. Just like sDDM it is defined by 4 parameters.
processes, they defined self-control as a value-based choice between two alter-
6Full DDM (fDDM) - Along with the 4 parameters of the sDDM it has an additional 4 natives. Rather than modeling self-control with dual-process models, they used
parameters : a standard deviation parameter characterizing the noise in the accumulation the drift diffusion model. By using the example of a dieter choosing between
process, a starting point bias parameter(zm), a range of latency times giving the distribution a salad and a burger, they looked as self-control as a comparison between the
from which the latency time (non-decision time) is sampled every trial and a range of bias subjective values of two alternatives, thereby eliminating the need for a ‘control’
giving the distribution of the bias parameter. system. The decision would be governed by the values of the decision thresholds
7Full Collapsing Barrier DDM (fcbDDM) - It is defined by the same 8 parameters as the
fDDM but now has collapsing barriers like the scbDDM. the ∆ BIC score, better the model. It penalizes the model for having too many observations
and parameters and rewards the model f or fitting the data well.
8Bayes Information Criterion - BIC is a criterion used to compare different models for a given set
of data samples. It strikes a balance between model complexity and model fit. Lower
96
95
and the two alternatives. Their model captures internal events like effort expen- (2020) [ZWB20], applied the DDM to this psychological phenomenon. They
diture by incorporating it into the value-integration process. Effort can be an modified the full DDM to incorporate the unequal weightage of losses against
opportunity cost that is compared with the benefits that an alternative pose., gains but also incorporated a pre-valuation bias. This bias behaved similarly to
thus the task that needs more effort can be avoided, favoring the impulsive the starting point bias, and represented a predisposition towards rejection, by
option over the deliberative option. This view of self-control could lead to un- being closer to the rejection bound (Figure 6). Since the starting point for the
derstanding why damage to prefrontal cortices, areas thought to participate in diffusion process is closer to the rejection bound, the decision-maker is biased
evidence accumulation and comparison of alternatives, results in more impulsive towards rejecting the gamble. It takes less evidence for the decision variable
decisions. It can also lead to further research into the realms of goal-attainment to cross the bound, thus the RTs for rejection will be shorter and the prob-
and motivation. ability of rejection will be greater. This bias corresponds to prior experience
and introduces the concept of learning into the experiment. As they show in
their paper, during trial blocks with higher payoffs –i.e. trials in which the
possible gains were much greater than the possible losses, this pre-valuation
bias was closer to the rejection bound, meaning it took a larger gain to loss
ratio to convince the subject to accept the gamble, in this case, 1.83, whereas
in trial blocks with lower payoffs i.e. trials in which the possible gains were
almost equal to the possible losses, the pre-valuation bias was farther from the
rejection bound, meaning that it took a smaller gain to loss ratio to convince
the subject to accept the gamble ,in this case, a 1.25 gain-to-loss ratio. Thus,
they show the influence of prior gambles and the prior rewards on the current
gamble. Using the Deviance Information Criterion, a model criterion similar
to BIC which penalizes the model with greater variance, and thus uncertainty
in the data, they showed how the DDM outperforms older models explaining
loss aversion. Through the incorporation of a starting point bias, in the form
of the pre-valuation bias, their model captures the choice probabilities and the
RTs for both rejected and accepted gambles. Thus, the DDM has been success-
Figure 5: Subjective value accumulates over time just as sensory in- fully modelled for another task, more cognitive than those of the economic and
formation does. The value of Action A accumulates rapidly but falls perceptual domains.
over after some time whereas the value for Action B rises slowly but
ultimately reaches a higher point. A person with a lower decision 5.3 Driving Tasks
threshold would pick Action A and could have poor self-control. Un-
der time pressure, A would be the action chosen. However, for a Recently, the DDM has been modified for different types of driving tasks.
person with a higher decision threshold or with no time pressure, B Cooper and Strayer, (2008) [CS08] conducted an experiment to determine the
would be chosen. This explanation can be heightened by taking the effects of cell-phone usage on driving. They used a 3D driving simulation during
example of Action A being eating pizza and Action B as eating salad. which subjects were engaging in a conversation they found interesting using a
(Figure adapted from Berkman et al., 2017 [BHL+ 17]) hands-free phone. Ratcliff and Strayer, (2014) [RS14] conducted an analysis on
this study using a one-boundary drift diffusion model (Figure 7A) and found
that distracted drivers have longer non-decision times and lower drift rates re-
sulting in longer response times and slower uptake of information. Thus, this
5.2 Loss Aversion study provided computational proof as to why distracted drivers have higher
Loss Aversion is one of the central tenets of Prospect Theory (Kahneman and chances of being in a car crash.
Tversky, 1979 [KT79]), which proposes that when faced with risk or uncertainty, Building on this, Daneshi et al., (2020) [DAT20] used a one-boundary DDM
decision-makers are loss averse i.e. they place a greater weight on losses than to model time-to-collision to an obstacle. In this task, the subjects had to stay
they do on gains. An experiment that highlights this is when decision-makers on their trajectory for as long as possible but prevent collision with the lead
are offered a gamble with a 50% probability to gain 11$ and a 50 % proba- vehicle. (Figure 7B). They conducted this task with and without time pressure
bility to lose 10$, they often reject the gamble. Despite the gamble having and found that both the drift rate and the decision threshold were higher for the
a positive expected value, it seems unattractive and is rejected. Zhao et al., trials with time pressure. This could mean that under time pressure drivers have
97 98
Figure 6: A drift diffusion process for loss aversion. The pre-valuation
bias γ appears as a starting point bias towards the rejection threshold.
Since the distance from the thresholds is now unequal, rejection is
more likely and will have a shorter response time since it takes less
evidence to reach the threshold. (Figure adapted from Zhao et al.,
2020 [ZWB20])
greater evidence accumulation but also can be uncertain about their decisions Figure 7: A) A one-boundary diffusion model for driving tasks. The pa-
thus increasing their decision thresholds and their margins of safety. rameters remain the same as the sDDM. Figure adapted from Ratcliff and
Both the previous studies have looked at simple braking and driving around Strayer, (2014) [RS14] B) Participants have to stick to the yellow line for
tasks. The DDM has also been applied to more complex tasks such as accepting as long as possible and need to drive around once the obstacle gets too
or rejecting a turn at an intersection. Zgonnikov and Abbink,(2020) [ZA20] close. Figure adapted from Daneshi et al., (2020) [DAT20] C) The partic-
used a modified full collapsing barrier DDM (fcbDDM) with variable drift rates ipants are in the red car. The speed of the blue oncoming car is variable.
to model a driving task which had subjects accept or reject a left turn with Participants have to decide whether or not to turn left. Figure adapted
an oncoming car which could block them (Figure 7C). Evidence accumulation from Zgonnikov and Abbink, (2020) [ZA20] D) The model predictions are
involved gauging the distance from the oncoming car and its speed and using this given by the dotted lines. They fit the data well. The probability to turn
information to compute a time-to-arrival. Greater the time-to-arrival, greater increases with an increase in time to arrival and an increase in distance
the probability to turn. They found a positive relationship between response from the oncoming car. The reaction time also shows a positive relation-
time and time-to-arrival (Figure 7D). Their model also accurately predicted ship with time-to-arrival. Figure adapted from Zgonnikov and Abbink,
their data well. This modification of a variable drift rate could be of huge (2020) [ZA20]
importance to similar experiments that look at dynamic real-life scenarios. More
research into driving behaviours could have applications in computer-driven cars
and in making traffic interactions safer.
5.4 Racial Bias

In light of the increasing police brutality, researchers have tried to study racial
biases in police officers and trainees and regular people. A first-person shooter
task (FPST) has been developed for these studies. In the task, the participants
are instructed to shoot armed targets and to not shoot unarmed targets. The
99 100
targets can be either Black or White men.
Using a hierarchical DDM 9 , Johnson et al., (2017) [JHCP17] analysed an
FPST in which participants were rewarded for correct shooting decisions. They
found that participants had a starting point bias towards the to-shoot decision,
however, this was independent of race and can be explained by the rewarding
outcome for to-shoot decisions. Evidence accumulation was stronger to shoot
armed Black men than to shoot armed White targets, thus participants have a
higher drift rate when it comes to shooting armed Black men and this results
in shorter response time and a greater likelihood to shoot armed Black men.
Following this study, Johnson et al., (2020) [JSCF20] looked at the effects of
sleep deprivation and caffeine on racial biases. They found that subjects were
more likely to shoot unarmed Black men than unarmed White men and this
bias was not affected by either sleep deprivation or caffeine. Caffeine did not
mitigate the errors caused by sleep deprivation. It only reduced response times.
They also found that subjects set a wider threshold for White men than for
Black men, showing that they needed lesser evidence when it came to making
a decision when they were shown a Black man as the target. Surprisingly, they
found that overall, participants who were given a placebo had a higher starting
point to shoot White targets. Figure 8: The x-axis shows the 4 groups of patients: Patients with a
This study reaffirmed a conclusion that Johnson et al., (2017) [JHCP17] had whole night’s sleep on a placebo, patients with a whole night’s sleep
come to, proving that for unarmed targets, subjects had a lower drift rate for on caffeine, patients who had not slept for 24 hours on placebo, and
Black men than for White men and for armed targets, had a higher drift rate patients who had not slept for 24 hours on caffeine. The top left
for Black men than for White men. panel shows that subjects had wider thresholds for white men than
for black men. The top right panel shows that subjects, surprisingly,
5.5 Reinforcement Learning had a starting point bias to shoot white men. The bottom left panel
shows the effect of sleep deprivation and caffeine. The bottom right
The DDMs looked at in the review so far have not incorporated an element panel shows the drift rates for the different trials. All the negative
of learning into the process, however, the drift diffusion process modeled for drift rates i.e. below the dashed lines are for unarmed targets while
loss aversion hinted at the influence of past outcomes. Recently, the DDM has those above the dashed lines are for armed targets. Thus, Black
been applied to learning tasks as well. These groups of models are called rein- armed men produced a higher drift rate in the participants while
forcement learning drift diffusion models (RLDDM). They unify the DDM and subjects had a lower drift rate for unarmed black men than those for
the theory of reinforcement learning (See Seo and Lee, (2012) for an excellent white men. Figure adapted from Johnson et al., (2020) [JSCF20]
review [SL12]). These models have been applied to probabilistic selection tasks
(PST), which present the participants with two options with the goal being to
pick the option with a greater probability to be rewarded. However, the partic- values, a scaling parameter vmod 11 , which ensures that the difference in values
ipants need to learn the probabilities and rewards of these options as the trials and probabilities for the two choices are transformed into an appropriate scale
go on. The simplest RLDDM has 4 parameters: a learning rate 10 , threshold in the DDM framework, and the non-decision time.
9 Hierarchical DDM – The HDDM analyzes the data at the population level rather than
Fontanesi et al., (2019) [FGSR19] showed that the RLDDM can explain
at an individual level. This means that fewer trials can be conducted per participant, but both reaction times and choice probabilities very well. However, it also shows
the parameters can easily be recovered and can still capture all the aspects of the data as the the learning throughout the task and thus successfully combined the RL models
simple DDM does. and the DDM (Figure 9).
10 Learning Rate - The learning rate determines how sensitive the decision maker is to The accuracy of the responses increased steadily and the RTs decreased
previous outcomes. A learning rate that is too low is not optimal since the learning will be throughout the task, showing that the participants learnt the probability of
very slow, however a learning rate that is too high will induce forgetting the outcomes that 11 Scaling Parameter - It is analogous to the drift rate in the DDM. It helps to convert the
happened a few trials back.
effect of previous outcomes into an appropriate scale that can be incorporated into the DDM
framework.
101 102
Figure 10: Lower the WAIC score, better the model. The dual learning
rate means that positive and negative outcomes will have a different
effect on the decision-maker. The threshold is variable, thus when
the decision is the easiest the threshold is the lowest. The scaling
Figure 9: The top panels show the power of the RL models whereas the parameter can be fixed or s-shaped.
bottom two represent the contribution of the DDM to the RLDDM.
Thus, two prominent theories can be unified to give a better account
chological phenomena and real-life decisions. The model captures the response
of decision making, a) The accuracy increases as the trial number
times and the behaviour of the participants in these tasks and helps in giving
increases thus showing the effect of learning on the task. b) The RT
a greater insight into the underlying processes of deliberation and decision-
decreases, once again showing the effect of learning. Figure adapted
making. Further research into the DDM could have powerful applications in
from Fontanesi et al., (2019) [FGSR19]
consumer behaviour, traffic behaviour and laws, computer-driven vehicles and
could have important clinical and social applications. Though the tasks looked
the rewards and improved their performance. They also conducted an analysis at in this review are more complex than sensory and simple value-based choice,
to find out which RLDDM explained the data the best. The RLDDM can be they are still a step away from explaining important and life-changing decisions.
modified by having different learning rates for negative and positive outcomes. The model comparison criteria used in this review (BIC, DIC, WAIC) may rep-
The threshold can either be fixed or variable, and the scaling parameter can resent a caveat in the literature. These criteria penalize the models in different
either be linear or sigmoid. Thus, there can be 8 types of the RLDDM. Using manners and lack of uniformity in the literature could result in the selection of
the Watanabe-Akaike Information Criterion 12 , they found that the full RLDDM incorrect models (Churchland and Kiani, 2016 [CK16]) . Future research should
i.e. with dual learning rates, one for positive and one for negative outcomes, aim to refine the DDM framework and attempt to resolve the debates about the
with sigmoid scaling parameters and with variable bounds explain the date the dynamics of the drift-diffusion process. Future work should also aim to conduct
best (Figure 10). more extensive research in finding definitive neural correlates and circuits for
Pedersen et al., (2017) [PFB17] used the RLDDM to gain a different per- the parameters. New techniques such as calcium imaging and optogenetics, if
spective on ADHD patients and the effects of medication. They found that adapted to work in primates, hold interesting possibilities. Another interesting
medication increased boundary separation, lowered learning rates, increased innovation in the literature is the quantum drift-diffusion model (Rosendahl et
non-decision time, and increased the drift rate scaling, showing the shift to- al., 2020 [RBC20]), which looks at evidence as a quantum particle of information
wards focusing on accuracy rather than speed. Thus, the RLDDM has the and the threshold as a square attractor. This may open new and fascinating
potential to be used in many clinical experiments. avenues for improving computational models as a whole. In the last 10 years,
the extensions of the Drift Diffusion Model have led to tremendous progress in
understanding how cognitive decisions are made. The DDM has the potential
6 Discussion to model more complex decisions and holds a lot of promise for the future.
Recently, researchers have tried applying the Drift Diffusion Model, a popular
computational model for sensory decision-making, to more cognitive and com- References
plex tasks. These studies have shown that the DDM can explain a variety of psy-
12 Watanabe Akaike Information Criterion - Another model criterion like BIC and DIC. It [BBHF10] U. Basten, G. Biele, H.R. Heekeren, and C.J. Fiebach. How the
penalizes the model in a way, similar to that of DIC but it takes the summation of the variance brain integrates costs and benefits during decision making. Proc.
of each posterior draw. It is more computationally taxing than both BIC and DIC but gives Natl. Acad. Sci. U. S. A., 2010.
a better approximation of how good the model is.
103 104
[BHL+ 17] E.T. Berkman, C.A. Hutcherson, J.L. Livingston, L.E. Kahn, and [HMU08] H.R. Heekeren, S Marrett, and L.G. Ungerleider. The neural sys-
M. Inzlicht. Self-Control as Value-Based Choice. Curr. Dir. Psy- tems that mediate human perceptual decision making. Nat. Rev.
hchol. Sci., 2017. Neurosci., 2008.
[BWFN10] R. Bogacz, E.J. Wagenmakers, B.U Forstmann, and S. Nieuwen- [HSC+ 11] T.A. Hare, W. Schultz, C.F. Camerer, J.P. O’Doherty, and
huis. The neural basis of the speed-accuracy tradeoff. Trends A. Rangel. Transformation of stimulus value signals into motor
Neurosci., 2010. commands during simple choice. Proc. Natl. Acad. Sci. U. S. A.,
2011.
[CK16] A.K. Churcland and R. Kiani. Three challenges for connecting
model to mechanism in decision-making. Curr. Opin. Behav. Sci., [IBM08] J. Ivanoff, P. Branning, and R. Marios. fmri evidence for a dual
2016. process account of the speed-accuracy tradeoff in decision-making.
PLoS One 3, 2008.
[CS08] J.M. Cooper and D.L. Strayer. Effects of simulator practice
and real-world experience on cell-phone-related driver distraction. [JHCP17] D.J. Johnson, C.J. Hopwood, . Cesario, and T.J. Pleskac. Ad-
Hum. Factors, 2008. vancing research on cognitive processes in social and personality
psychology: A hierarchical drift diffusion model prime. Soc. Psy-
[DAT20] A. Daneshi, H. Azarnoush, and F. Towhidkhah. A one-boundary chol. Personal. Sci, 2017.
drift-diffusion model for time to collision estimation in a simple
driving task. J. Cogn. Psychol., 2020. [JSCF20] D.J Johnson, M.E. Stepan, J. Cesario, and K.M. Fenn. Sleep
deprivation and racial bias in the decision to shoot: A diffusion
[FBD+ 10] B.U. Forstmann, S. Brown, G. Dutilh, J. Neumann, and E.J. Wa- model analysis. Soc. Psychol. Personal. Sci, 2020.
genmakers. The neural substrate of prior information in perceptual
decision making: A model-based analysis. Front. Hum. Neurosci., [KCRN14] R. Kiani, C.J. Cueva, J.B. Reppas, and W.T. Newsome. Dynam-
2010. ics of neural population responses in prefrontal cortex indicate
changes of mind on single trials. Curr. Biol., 2014.
[FDB+ 08] B.U. Forstmann, G. Dutilh, S. Brown, J. Neumann, D.Y. von
[KLCR12] I. Krajbich, D. Lu, C. Camerer, and A. Rangel. The atten-
Cramon, K.R. Ridderinkhof, and E.J. Wagenmakers. Striatum
tional drift-diffusion model extends to simple purchasing decisions.
and pre-sma facilitate decision-making under time pressure. Proc.
Front. Psychol., 2012.
Natl. Acad. Sci. U. S. A., 2008.
[KR11] I. Krajbich and A. Rangel. Multialternative drift-diffusion model
[FGSR19] L. Fontanesi, S. Gluth, M.S Spektor, and J. Rieskamp. A reinforce-
predicts the relationship between visual fixations and choice in
ment learning diffusion decision model for value-based decisions.
value-based decisions. Proc. Natl. Acad. Sci. U. S. A., 2011.
Psychon. Bull. Rev., 2019.
[KT79] D. Kahneman and A. Tversky. Prospect theory: An analysis of
[FRW16] B.U. Forstmann, R. Ratcliff, and E.J. Wagenmakers. Sequential decision under risk. Econometrica, 1979.
sampling models in cognitive neuroscience: Advantages, applica-
tions, and extensions. Annu. Rev. Psychol., 2016. [LP11] T. Liu and T.J. Pleskac. Neural correlates of evidence accumula-
tion in a perceptual decision task. J. Neurophysiol., 2011.
[GS07] J.I. Gold and M.N. Shadlen. The Neural Basis of Decision Making.
Annu. Rev. Psychol., 2007. [MMH+ 10] M. Milosavljevic, J. Malmaud, A. Huth, C. Koch, and A. Rangel.
The drift diffusion model can account for the accuracy and reac-
[HBS09] T.C. Ho, S. Brown, and J.T. Serences. Domain general mecha- tion time of value-based choices under high and low time pressure.
nisms of perceptual decision making in human cortex. J. Neu- Judgm. Decis. Mak, 2010.
rosci., 2009.
[MvMF14] Mulder.M.J, L. van Maanen, and B.U. Forstmann. Perceptual
[HKS+ 12] L.T. Hunt, N. Kolling, A. Soltani, M.W. Woolrich, M.F.S Rush- decision neurosciences - a model-based review. Neuroscience, 2014.
worth, and T.E.J. Behrens. Mechanisms underlying cortical ac-
tivity during value-guided choice. Nat. Neurosci., 2012. [MWR+ 12] M.J. Mulder, E.J. Wagenmakers, R. Ratcliff, W. Boekel, and B.U.
Forstmann. Bias in the brain: A diffusion model analysis of prior
probability and potential payoff. J. Neurosci., 2012.
105 106
[PAHB11] M.G. Philiastides, R. Auksztulewicz, H.R. Heekeren, and [VVKC08] V. Van Veen, M.K. Krug, and C.S. Carter. The Neural and Com-
F. Blankenburg. Causal role of dorsolateral prefrontal cortex in putational Basis of Controlled Speed-Accuracy Tradeoff during
human perceptual decision making. Curr. Biol., 2011. Task Performance. J. Cog. Neurosci., 2008.
[PFB17] M.L. Pedersen, M.J. Frank, and G. Biele. The drift diffusion model [WvMR+ 12] K. Winkel, L. van Maanen, R. Ratcliff, M.E. Van der Schaaf, M.R.
as the choice rule in reinforcement learning. Psychon. Bull. Rev., Van Schouwenburg, R. Cools, and B.U. Forstmann. Bromocriptine
2017. does not alter speed-accuracy tradeoff. Front. Neurosci., 2012.
[PSA06] C. Padoa-Schioppa and K=J.A. Assad. Neurons in the or- [ZA20] A. Zgonnikov and D. Abbink. Should i Stay or Should I
bitofrontal cortex encode economic value. Nature, 2006. go? Evidence Accumulation Drives Decision Making in Drivers.
PsyArXiv, 2020.
[Rat78] R. Ratcliff. A Theory of Memory Retrieval. Psychol. Rev., 1978.
[Rat04] R. Ratcliff. A Comparison of Sequential Sampling Models for [ZWB20] W.J. Zhao, L. Walasek, and S. Bhatia. Psychological mechanisms
Two-Choice Reaction Time. Psychol. Rev., 2004. of loss aversion: A drift-diffusion decomposition. Cogn. Psychol.,
2020.
[RBC20] M. Rosendahl, A. Bizyaeva, and J. Cohen. A Novel Quantum
Approach to the Dynamics of Decision Making. Cognitive Science
Society, 2020.
[RGD10] E.T. Rolls, F. Grabenhorst, and G. Deco. Decision-making, errors,
and confidence in the brain. J. Neurophysiol., 2010.
[RM11] R. Ratcliff and G. Mckoon. Effects of aging and iq on item and
associative memory. Pyschol. Rev., 2011.
[RMG04] R. Ratcliff, G. McKoon, and P. Gomez. A diffusion model account
of the lexical decision task. Psychol. Rev., 2004.
[RS02] J.D. Roitman and M.N. Shadlen. Response of Neurons in the Lat-
eral Intraparietal Area during a Combined Visual Discrimination
Reaction Time Task. J. Neurosci., 2002.
[RS14] R. Ratcliff and D. Strayer. Modeling simple driving tasks with a
one-boundary diffusion model. Psychon. Bull. Rev.w, 2014.
[RTM10] R. Ratcliff, A. Thapar, and G. Mckoon. Individual differences ,
aging , and iq in two-choice tasks. Cogn. Psychol., 2010.
[SK08] C. Summerfield and E. Koechlin. A neural representation of prior
information during perceptual inference. Neuron, 2008.
[SL12] H. Seo and D. Lee. Neural basis of learning and preference during
social decision-making. Curr. Opin. Neurobiol,, 2012.
[TLvKV20] M. Theisen, V. Lerche, M. von Krause, and A. Voss. Age differ-
ences in diffusion model parameters: a meta-analysis. Pyschol.
Rev., 2020.
[vBM15] M. von Boguslawski and P. Mildén. The attentional drift-diffusion
model for simple choice in the quaternary case, measuring the
effect of permutation of item location on choice. Behaviour, 2015.
107 108
Procrastination has long been a problem in society [FE95] [Mil08] and has
recently gained significant attention in the field of clinical psychology. Glob-
ally, about 20-25% of adults are chronic procrastinators in a variety of life sit-
uations like academic pursuits, social relationships, professional settings, and
Anxiety and Procrastination: What is the finance management [BD07] [FDM14] [KC19]. Being one of the most prevalent
Association? problems among students, procrastination is found to be the major barrier in
learning by one-third of the general population [SF13]; 95% of American col-
∗
lege students reported engaging in academic procrastination, with almost half
Xiaoyu Wu of them procrastinating on at least 50% of tasks [EK77] [BD07].
In fact, despite its high prevalence, procrastination is indeed a multifactorial
April 2, 2021 psychological phenomenon that involves a complex interaction of behavioral,
cognitive, and affective components [SR84]. It is identified as one of the least
understood human behaviors, yet it leads to not only lower levels of wealth,
health, and well-being, but also psychological distress [BD07] [SF13] [AA15].
Abstract
More specifically, procrastination is closely connected to poor academic per-
Psychologically, procrastination is understood as a form of behavioral formance [PMM+ 17] [GH19], reduction in work productivity [FBN+ 15], neg-
self-handicapping whereby an individual delays beginning or completing ative emotional and behavioral reactions [SWS00] [FDM14], and more acute
a task to strategically avoid situations that may show an adverse image. health problems [Sir07]. [SM13] even noted the fact that procrastinators make
Given the high prevalence of procrastination in our daily lives, recent work
more errors, work slower, and miss more deadlines when compared to non-
has begun to investigate its connection with other psychological factors
procrastinators. Although most procrastinators consider their behaviors as in-
and psychopathology. Both theoretical models and people’s first-hand
experience have indicated a possible association between procrastination appropriate, problematic, and in need of change [SM13] [KC19], it is quite
and anxiety. This review summarizes the existing literature and integrates possible that they do not have much insight into the psychological mechanisms
findings within the conceptualization of procrastination. Anxiety and proof why they repeatedly engage in such undesirable behavior.
crastination are concluded to be closely correlated and possible psycho- One psychological factor that may be linked to procrastination is anxiety,
logical mechanisms that help explain such correlation are proposed. It is another phenomenon with profound effects on the global population. According
speculated that anxiety can be both the result and the driver of procras- to the World Health Organization (2017), 264 million adults around the world
tination, and procrastination can be a strong predictor of psychological have anxiety and it is estimated that 31.3% of all U.S. adults will experience
conditions. These findings will contribute to a better understanding of an anxiety disorder, the most common mental illness in the U.S., at some point
procrastination and its psychological correlates, providing crucial impli- in their lives [Sch07]. Large population-based surveys suggest that up to 33.7%
cations for people who suffer from procrastination as well as for clinicians
of the population are affected by an anxiety disorder during their lifetime, and
who are trying to help chronic procrastinators. Future research is still
there is a substantial under-recognition and under-treatment of these disorders
needed to further confirm the direction of the causal relationship, to de-
termine the mediators and moderators of the relation, and to specify the [BM15]. For example, for students who worry that they are inadequate to pass
clinical value of the conclusion. their classes, anxiety may force them to drop out of college; for adults, they
carry such anxiety into life even after they graduate and it will continue to
affect their work and social lives [CDMM05] [Mil08]. Though anxiety cannot
1 Introduction kill people directly, more severe forms of anxiety can result in people committing
suicide as an alternative to the suffering [Mil08]. As only 9.8% of those suffering
Simply defined, procrastination refers to the act of “put[putting] off intentionally received possibly adequate treatment globally [ALEL+ 18], further research on
the doing of something that should be done” [cdnd]. Psychologically, procras- anxiety, its related symptoms, and applicable treatments is crucial.
tination is understood as a form of behavioral self-handicapping whereby an Based on the perspective of seeing procrastination as a manifestation of
individual delays beginning or completing a task to strategically avoid situa- lack of self-control and time-management skills, current mainstream treatments
tions that may show an adverse negative image [FT00] [M.04]. Importantly, for procrastination, such as Cognitive Behavioral Therapy (CBT), are rather
in addition to its behavioral components, procrastination also entails emotional more behavioral or developed from a motivational or volitional standpoint [B.13]
and cognitive components [Fer1b] [RSM86] [M.04] [RBF+ 18]. In particular, CBT for procrastination mainly targets improving self-
∗ Advised by: Ema Tanovic of Yale University
regulation skills, goal-setting techniques, implementation intentions, and time
management ability [B.13] [RBF+18], and usually involves creating a prioritized
109 110
to-do list or even increasing the perceived pressure. In general, it is signifi- tional beliefs; first, procrastinators hold poor competence beliefs, and second,
cantly different from the more well-known CBT for anxiety as relatively little they are fearful of the possible negative social consequences of failing to complete
attention is given to anxious thoughts, emotions, and physiological sensations. the task well enough [FBN+ 16]. Such fear of failure has also been discovered in
A recent meta-analysis showed that the benefit of current psychological treat- empirical studies on procrastination; they have discovered not only a positive
ments for procrastination is minimal and the effectiveness of CBT specifically is correlation but also a causal relationship between anxiety and procrastination,
not as satisfactory as conventionally expected [RBF+ 18]. Although new forms in which fear of failure would lead to a higher level of procrastination [ZDF+ 18].
and types of treatments like internet-based interventions [KAEB19], watching Particular cognitive factors are further revealed in other studies. For example,
videos [BSPT20], and acceptance-based behavioral therapy (ACT) [GO15] are task aversiveness, the extent to which a task or behavior is perceived to be un-
being developed, there is still a lack of effective and personalized treatments of pleasant or difficult to perform by an individual [BP00], is proved to be closely
procrastination. Thus, further research on the mechanisms and causal factors related to procrastination [SR84]. Such aversiveness, or unpleasant feeling, can
involved in procrastination is important. be seen as a result of excessive anxiety or fear of failure, as people clearly
Previous research on the topic of procrastination has mostly focused on its would not enjoy performing a task when overwhelmed with worries about neg-
outcomes and interventions instead of the causes. For instance, [LLSN18] spec- ative future outcomes. A few studies further imply a close correlation between
ify the negative influence of leader procrastination on group functioning at work, stress and task aversiveness as well as between task apprehension and fear of
and [vE03] examines the effectiveness of time management on treating procras- failure [BP00] [OC01]. Therefore, it is possible that as one is holding such a neg-
tination. In addition, previous studies mainly investigate forms of procrastina- ative attitude towards future tasks and himself, one then experiences excessive
tion that are related to academic scores [ZH18], work performance [MPT18], stress, apprehension, and in other words, anxiety; then, such anxiety may lead
internet use like video-game addiction [YWH+ 17], and bedtime delay [KSD18]. to procrastination, being a mediator between irrational or negative cognitive
Furthermore, existing studies intensively focus on the cognitive processes of beliefs and procrastination.
procrastination [Fer1b], leaving out the characterization of behavioral, arousal, In addition, from a more behavioral perspective, some psychologists have
affective and emotional factors like anxiety. Therefore, the relationship between viewed procrastination as an avoidance behavior that arises in response to anx-
anxiety and procrastination remains unclear. iety. Procrastinators are found to avoid situations that may reveal information
The current review examines the association between anxiety and procras- concerning their abilities [Fer1b] [RF20]. Procrastination is proposed to be
tination, whether anxiety and procrastination are related, whether anxiety is closely related to short-term mood repair and emotion regulation [SP13]. Such
the result or the driver of procrastination, and whether procrastination can be findings have led theorists to suggest that procrastination is a type of avoid-
a sign or predictor of psychopathology by reviewing evidence on the correlation ance whereby people procrastinate to prevent themselves from experiencing the
between anxiety and procrastination, a potential causal relationship in both di- anxiety of doing the task or confronting the idea that they might fail. The
rections, as well as moderators of the relation. Elucidating the role of anxiety appraisal-anxiety avoidance (AAA) model states that avoidance, a behavioral
in procrastination will contribute to a better understanding of the psychological response to stress, functions to reduce the perceived anxiety when people find
mechanism between anxiety and procrastination and provide crucial implica- themselves inadequately prepapred to cope with the threat. As avoidance is
tions for people who procrastinate as well as for clinicians who are trying to so effective at relieving immediate anxiety, it ultimately perpetuates the task
help those chronic procrastinators. Furthermore, specifying whether anxiety is avoidant pattern or the behavior of procrastinating, as well as reinforces anx-
a key causal factor of procrastination will help inform the direction of future iety [MT99]. In particular, some refer to procrastination as the manifestation
development of treatment and thus help reduce negative outcomes. of defensive avoidance, the attempt to avoid or postpone the stress of being
exposed to relevant information of the decision or task [Eva90]. In a similar
1.1 Possible Mechanisms Linking Anxiety and Procrasti- manner, Mowrer’s two-factor model of anxiety [Mow47] highlights the impor-
tant role of negative reinforcement and is consistent with avoidance behavior as
nation
a maintaining factor of procrastination [FE95]. Although this perspective may
Several existing models of procrastination do indicate a potential relationship fail to consider individual differences among procrastinators [FE95], it does
between anxiety and procrastination, where anxiety might be a key factor in provide a powerful demonstration of the possible relation between anxiety and
explaining procrastination. Researchers have commonly viewed procrastination procrastination.
as a self-regulatory or cognitive failure [BHT94] [TS89] [FBN+ 16] and an il- Irrational or negative thoughts resulting from a biased time perspective
logical and non-goal directed behavior where irrational cognitions play a key imply psychological mechanisms between anxiety and procrastination as well.
role [EK77]. This could imply possible mechanisms of the relation between Procrastinators are found to have the tendency of focusing more on past and
anxiety and procrastination. From the Rationale-Emotive Behavioral Therapy present events as compared to the future, a cognitive orientation that is re-
(REBT) perspective, there are two assumptions about procrastinators’ irra- ferred to as time perspective [Fer91] [RF20] [M.14]. It is possible that such a
111 112
past-oriented thinking style would lead to thought control problems like rumi- crastination are still lacking.
nation, which is defined as a kind of intrusion that involves “repetitive, pro-
longed, and recurrent thought about one’s self, one’s concerns and one’s experi-
ences” [SLF00] [FSH+ 12]. It is suggested that procrastination is associated with 2 Literature Review and Analysis
a higher frequency of intrusive thoughts like rumination [RRBvdL18]. There-
fore, we could speculate that it is those overwhelming negative thoughts that 2.1 Evidence of a Correlation Between Anxiety and Pro-
then lead to excessive anxiety, which in turn results in procrastination. Indeed, crastination
individuals with anxiety disorders are also shown to have a negative past time
As researchers have begun to investigate the role of anxiety in the development
perspective, which is further related to rumination [ASWC18]; procrastinators
of procrastination and to measure both procrastination and level of anxiety
might be experiencing the same cognitive bias and thus, anxiety as well.
at the same time in recent years, there is more evidence showing a positive
On the other hand, although focusing more on the future orientation, worry
correlation between anxiety and procrastination. In general, a representative
could be another psychological pathway between anxiety and procrastination.
study conducted by Beutel and colleagues (2016) [BMA+ 16] showed that pro-
Worry is understood as a cognitive phenomenon that occurred when the individ-
crastination is consistently associated with higher stress and anxiety. The study
ual experiences a threat concerning possible future events and is often accom-
examined over 2,500 participants who were between the ages of 14 and 95 years
panied by feelings of anxiety [DGL01] [MWB91] [SJ01]. Both theoretically and
and demonstrated evidence of a strong correlation between anxiety and pro-
empirically, worry shows a substantial relationship with procrastination and a
crastination across the lifespan. Many other researchers have demonstrated the
significant positive causal effect on anxiety [SJ01] [GMC01].
same result as well [CBM04] [KC19] [VFGI12] [HMS98]. Many studies have been
Another model that helps explain the correlation between anxiety and pro-
done in academic settings and specifically examine academic anxiety and pro-
crastination is the utility expectancy-value theory [Ecc83] [dJEW02]. Central
crastination. For example, Custer (2018) [Cus18] administered the Test Anxiety
to the theory are two concepts: first, the utility value, which is defined as the
Inventory and the Procrastination Assessment Scale for Students to over two
perceived usefulness of a particular task or activity in achieving goals [SJ19],
hundred prelicensure nursing students, aged 19 to 53, in America, and found a
and second, the expectancy, a measure of the extent to which an individual be-
statistically significant correlation between test anxiety and academic procras-
lieves that a given task will yield utility [FBN+ 17]. The perceived utility value
tination. Similarly, through self-report measures and a canonical correlation
is found to be closely related to avoidance intentions and procrastination [SJ19];
analysis, Vahedi and colleagues (2012) [VFGI12] found a positive association
instead of procrastinating on tasks labeled as fun or pleasurable, procrastina-
between a latent statistics anxiety factor and procrastination among Iranian
tors only procrastinate when the task was identified as evaluative – or in other
undergraduate students. It is important to note that both of these studies were
words, valuable [FT00]. It might be explained by the fact that individuals may
conducted in mostly female samples, which may limit the generalizability of
feel a greater emotional burden and higher anxiety when participating in more
their findings.
important and useful tasks [NLL11] [SBE11] [BGP17]. Those two factors that
The relation between anxiety and academic procrastination has also been ex-
motivate people’s achievement-related choices can be reasonably tapped into
amined in younger populations. In a sample of younger participants (aged 13,
several cognitive constructs that have been implicated in procrastination and
14, and 16) in Israel, it was also found that students who are under higher levels
relevant to anxiety, namely fear of failure, task aversiveness and worry. Arguably
of anxiety are prone to procrastinate more on assignments like preparing for ex-
people may choose to procrastinate because the subjective high importance of
aminations and writing papers than those who are having less anxiety [MT99].
the task makes them feel excessively anxious.
Similarly, Rosário and colleagues (2008) [RNS+ 08] conducted a study in two
Recent research has begun to investigate the neural substrates of procras-
different samples, which consisted of over a thousand participants in total, of
tination [ZCX+ 20]. This type of research may be informative if similar neural
junior-high students in Portugal, and discovered a positive and significant cor-
circuits are implicated in procrastination compared to those implicated in the
relation between test anxiety and procrastination. Overall, it is evident that the
experience of anxiety. Research has shown that among chronic procrastina-
correlation between anxiety and academic procrastination exists among people
tors, the volume of the amygdala is larger than normal and that the amygdala
of different ethnicities, living environments, and ages. Especially for academic
is less connected to the dorsal anterior cingulate cortex, which helps to regu-
procrastination, such a correlation between anxiety and procrastination could
late the amygdala’s reactions [SFP+ 18]. The amygdala is also a key element
be better understood by viewing it from the REBT perspective and noting the
of the anxiety circuitry and is responsible for fear and anxiety-related behav-
role fear of failure has played. Students who spend a great amount of time and
iors [BPCKB18]. Some work further suggests specific structures in the brain,
effort preparing for an exam or finish their papers on time and properly yet re-
specifally the right hippocampus, that could be account for the link between
ceive poor scores or evaluations are forced to shamefully acknowledge that they
anxiety and procrastination [ZCX+ 20]. However, work in this area is currently
are deficient i n i ntellectual abilities compared t o t heir counterparts. Therefore,
limited, and s pecific hypotheses about which r egions may be i nvolved i n pro-
113 114
in this case, procrastination, being a strategic cognitive choice, is adopted by to avoid certain tasks or behaviors to reduce the anxiety that comes from their
those who are anxious due to the deep fear of their academic inadequacy and perceived inadequacy to perform well. The results indicate that the correlation
consequent failure when facing an assignment. Such a cognitive process is very may apply to various types of anxiety and to different social settings, especially
common and shared by all groups of students. those that involve a self-appraisal process of individual skills or abilities.
Some studies have shown a more multifactorial and complex correlation be- In addition, other forms of procrastination like bedtime procrastination,
tween anxiety and academic procrastination. In a sample of American college which refers to the delay of bedtime with no external reasons, and workplace
students, a more complicated correlation is demonstrated: procrastinators ex- procrastination are both proposed to be closely and positively related to anxi-
perience less stress and anxiety early on when they procrastinate than do non- ety [CAS20] [PAZ18]. Further, procrastination may occur not only at the level
procrastinators, and there even appears to be a negative correlation between of the individual but also with groups. Hooft and Mierlo (2018) [vHVM18] in-
anxiety and procrastination earlier in the semester [TB97]. However, later on in vestigate team-level procrastination and arrive at a similar result, pointing out
the semester and in the overall process, procrastinators experience more stress that team procrastination is closely connected to increased stress levels among
and anxiety compared to those who do not procrastinate [TB97]. This can the members. Evidence of the association between anxiety and procrastina-
be explained by the two-factor model in which procrastinators procrastinate to tion in more diverse social settings is still necessary, but overall, there is strong
reduce their level of stress and anxiety, and the result in fact may be seen as em- evidence of a correlation between procrastination and anxiety across studies.
pirical evidence for the hypothesis that anxiety and procrastination are related Additionally, researchers have begun to investigate the correlation from a
through negative reinforcement. Although the behavior has been repetitively neurobiological perspective. Zhang and colleagues (2019) [ZLF19] revealed that
reinforced, when the deadline approaches, students are no longer allowed to put individual differences in procrastination can be attributed to “structural abnor-
things off; therefore, the anxiety comes back and even becomes more intense malities and altered spontaneous metabolism in the parahippocampal cortex
since the reduction effect brought by procrastination no longer exists and most and the prefrontal cortex” (pp. 817–830), which are regions involved in thinking
procrastinators do not master an alternative coping strategy. Therefore, timing about the future and emotion regulation, respectively. This hints at how neural
and external factors like task requirements may also affect the relation between correlates of procrastination could help explain some psychological correlates
anxiety and procrastination. of procrastination, particularly time perspective and anxiety. Furthermore, a
A few researchers have come to a somewhat different conclusion on the cor- pilot study done by Zhang and colleagues (2020) [ZCX+ 20] not only confirmed
relation between anxiety and procrastination. In a study done by Milgram the existence of the correlation between anxiety and procrastination but also
and Toubiana (1999) [MT99], when it comes to assignments like homework, the highlighted possible neurobiological evidence for an association between anxi-
relation is reversed, in which students who were more anxious about their home- ety and procrastination. They found that anxiety and trait procrastination are
work completed it more quickly than those who were less anxious. The authors each associated with the activity of the right hippocampus through conjunction
speculate that homework involves less task-centered anxiety and consequence- analysis and pointed out a positive correlation between the right hippocampal
centered anxiety, which may be different from the anxiety students experienced grey matter volumes and trait anxiety, as well as procrastination. Although
when facing exams and writing tasks. In fact, this can be understood as an ap- the study examined over 200 participants and provides compelling evidence,
plication of the utility expectancy-value theory, in which the perceived value of more biological studies on the correlation between anxiety and procrastination
completing daily homework may be lower than preparing for exams or finishing is needed to help answer the question of why exactly the correlation exists and
a paper and students may feel less anxiety and burdened to do it. However, the to solidify or falsify existing theories.
finding may also imply that types of anxiety, the severity of anxiety, or other
moderators can play a key role in the relation. Mixed results on this correlation 2.2 Evidence of Anxiety Leading to Procrastination
indicate a need for studies to further examine different possible factors.
Beyond the academic domain, a correlation is also found under other circum- There has been evidence showing a direct causal relationship between anxiety
stances that involve other forms of anxiety or procrastination. In particular, and procrastination, in which higher levels of anxiety lead to procrastination.
Phillips and colleagues (2015) [PtDA15] examine how interpersonal skills anxi- By recording the daily affect and events each participant experienced for two
ety, the anxiety one perceives due to the extent to which they believe in their weeks, [PH20] specified the direction of the relationship between procrastination
capability to interact and communicate with others, relates to procrastination. and negative affect (NA), which mainly include stress and anxiety. Noting
They conclude that there is a positive association between interpersonal skills that people reported more frequent procrastination following the days that they
anxiety and procrastination. In the same study, ”avoidance of help-seeking,” experience higher levels of NA and implementing a multilevel regression model,
another form of maladaptive avoidance behavior, is also found to be positively they found that NA predicted next-day procrastination and concluded that
related to anxiety factor while being negatively related to interpersonal skills. negative emotion is the motivator of procrastination behavior. Similarly, by
Both of t he findings are consistent with t he AAA model, i n which people t ry using a structural equation model, Paechter and colleagues (2017) [PMM+17]
115 116
revealed that statistics anxiety led to higher procrastination in a larger sample. and procrastination.
Though such rigorous calculations should be done in more diverse samples, Other findings have also underlined the key role of time perspective in con-
there is already some evidence that indicates a direct causal effect of anxiety on necting anxiety and procrastination together and explaining the effect of other
procrastination. psychological factors. For instance, it is found that lower resilience leads to
On the contrary, a few researchers have arrived at a different conclusion procrastination both directly and indirectly; in particular, social anxiety serves
that does not support such a causal effect of anxiety on procrastination. For ex- as a partial mediator in the negative relationship between resilience and pro-
ample, Rabin and colleagues (2011) [RFNU11] demonstrated that anxiety was crastination, as shown in a structural equation model analysis [KC19]. In other
not a predictor of procrastination when including demographic variables like words, both resilience, the individual capacity to overcome or to adapt to ad-
age, sex and ethnicity, a number of medical and psychiatric diseases or condi- versity through social interactions [Che14], and social anxiety, which is closely
tions, as well as additional variables like estimated IQ, depressive symptoms, related to negative beliefs about one’s self, are highlighted to be important con-
neuroticism, and conscientiousness in the linear regression models. Similarly, tributors to the development of procrastination. For possible explanations of
although they confirmed the positive correlation between anxiety and procras- those findings, according to [Fer1a], procrastinators tend to experience greater
tination, Haycock and colleagues (1998) [HMS98] also found that anxiety was public self-consciousness and social anxiety; further, it is also known that peo-
not a predictor of procrastination when entering variables including gender, age, ple with anxiety disorders are much more inclined to have negative past time
efficacy expectations, and anxiety into the regression model. In particular, both perspectives and therefore experience more repetitive negative thinking like ru-
depression and self-efficacy beliefs may be closely related to anxiety as well as mination [ASWC18]. Such a biased time perspective makes people dwell on the
procrastination [KBM07] [NL20] [BMA+ 16]. This suggests that there may not past and continuously ruminate on themselves and the events that have already
be a unique relationship between anxiety and procrastination and highlights happened, particularly focusing on the negative aspect; those people result in
the need for examining the potential role of other variables, like depression and possessing negative self-efficacy beliefs about themselves and thus experienc-
efficacy expectations, simultaneously and interpreting anxiety in the context of ing excessive anxiety when making social interactions. Further, according to
its relationship with those variables. Vassilopoulos and Watkins (2009) [VW09], rumination indeed maintains those
negative beliefs among those with social anxiety. Therefore, the mediation role
2.2.1 Evidence of Anxiety as a Mediator Between Procrastination of social anxiety here may be understood as a consequence of procrastinators’
and Other Factors past-focused mode of thinking. In general, viewing anxiety as a mediator in
the relationship between time perspective and procrastination provides critical
Currently, there is relatively more research that characterizes anxiety as a me- insights for understanding the relationship between anxiety and procrastination.
diator between procrastination and other psychological factors, but the results In addition, anxiety is also found to be a mediator of the relationship be-
are mixed. One of the most promising findings is the mediating role of stress tween certain cognitive beliefs and procrastination. In particular, [DPMM+ 17]
and anxiety in the relationship between time perspective and procrastination. investigated how positive and negative metacognitive beliefs about procrastina-
A meta-analysis that contained over four thousand participants showed that tion influence decisional procrastination. They concluded that anxiety partially
procrastination is negatively correlated with future time perspective yet pos- mediated the relationship between positive beliefs and procrastination and fully
itively correlated with present time perspective [M.14]. Further, it is demon- mediated the relationship between negative beliefs and procrastination; more-
strated that stress and positive affect (PA) partially mediated the relationship over, both negative and positive beliefs predicted higher levels of anxiety. Put
between future time perspective and procrastination [M.14]. It is particularly more simply, the cognitive perception of procrastination as a useful coping strat-
noteworthy that stress was negatively correlated with future time perspective egy or as an uncontrollable tendency to delay may indeed contribute to a higher
and positively associated with procrastination [M.14]. These results suggest level of worry about one’s own performance. In this case, it can be speculated
that procrastinators are cognitively biased to focus less on the future, and that that such engagement in the maladaptive metacognitions may occupy many
this is partially due to stress. It is clear that stress is closely connected to the mental assets that should have been responsible for initiating or completing
level of anxiety since anxiety can actually be seen as a reaction or an integral tasks and reinforce negative self-efficacy beliefs by causing more worries. Anxi-
of stress [AoAAnd]. Further, it is found that anxiety symptoms are closely as- ety, consisting of negative beliefs, thoughts and emotions about one’s abilities,
sociated with negative past and fatalistic present time perspective, rather than intelligence or the likelihood of success [WM96] [DPMM+ 17], is thus experi-
a future perspective [KLMSD+ 19]. Therefore, it can be concluded that people enced and then results in the tendency of escaping from or postponing task
with a future time perspective probably experience less anxiety and thus less execution.
procrastination. These results are highly consistent with the idea of viewing Similarly, [CBM04] examined how anxiety might mediate the relation be-
procrastination in the context of individual time perspectives, implying a pos- tween locus of control and procrastination. Locus of control refers to the extent
sible mechanism or a more fundamental cause of the relation between anxiety to which an individual believes that they have control over life events; under
117 118
this scenario, it reflects how students perceive the causes of their academic suc- mediators in the relationship between anxiety and procrastination, it is proba-
cess or failure [CBM04]. It is found that students who are internally oriented ble that when people are under excessive anxiety or other unpleasant emotions,
experience less procrastination and debilitating test anxiety compared to stu- they think much more irrationally and pessimistically to avoid upward counter-
dents who are externally oriented [CBM04]. The result appears to be logical factuals and prefer downward counterfactuals, which can be seen as an attempt
since students who believe in the strong connection between behavior and con- to restore positive mood through escaping the unpleasant state and avoiding
sequences and the effectiveness of exam preparation would feel less anxious and stressors. Such maladaptive cognition then leads to procrastination.
then procrastinate less compared to those who believe in luck, fate and chance.
These results support the conceptualization of procrastination as a cognitive 2.2.3 Evidence of Moderators of the Relation Between Anxiety and
self-regulatory failure, resulted from certain maladaptive cognitions. After all, Procrastination
anxiety may serve as a key link that connects some of the people’s fundamental
cognitions to the induction of certain behaviors like procrastination. Several moderators of the relation between anxiety and procrastination have
One promising area for future research on anxiety as a mediator is spec- been implied. One of the most supported moderators is future time perspective.
ifying its role in the relationship between life satisfaction or other individual Empirically, it is suggested that future time perspective is negatively correlated
socio-economic conditions and procrastination. Beutel and colleagues (2016) with effort cost, avoidance intentions, and procrastination; it may buffer the
[BMA+ 16] pointed out that procrastination is negatively correlated with overall causal effect of anxiety on procrastination by reducing effort cost and avoidance
life satisfaction and that specific individual conditions like lack of a partnership intentions [SJ19]. Theoretically, such speculation is consistent with the time
and unemployment are all predictors of procrastination. It can be speculated perspective theory: people with a future time perspective may experience less
that lower life satisfaction that may be due to any of those predictive measures rumination and put more emphasis on the big picture or long-term goals. Over-
may lead to procrastination by causing excessive anxiety or perceived stress. all, time perspective could be an important moderator in the relation between
Such a conclusion is not well-supported and few studies have examined this me- anxiety and procrastination.
diation role of anxiety in the relationship between individual living status and There is also evidence suggesting a moderating effect of age on the relation
procrastination, but overall, there is now evidence suggesting an important role between anxiety and procrastination. In general, the research on the relation
of life-condition-related anxiety and stress, having particularly large social and between age and procrastination are inconclusive [RFNU11], with some studies
clinical implications. reporting negative correlations [BRM88] [PMAP00] [vE03] and others reporting
no meaningful correlations [HMS98] [HWPB06]. In a recent study, [BMA+ 16]
assessed people’s level of procrastination across life domains and demonstrated
2.2.2 Evidence of Mediators Between Anxiety and Procrastination
that lower age is a predictor of procrastination through multivariate analysis.
In addition to causing procrastination directly or serving as a mediator in other Additionally, [RFNU11] found that increased age was a significant predictor
associations with procrastination, anxiety is also hypothesized to cause procras- of academic procrastination, which is consistent with the prior result. Other
tination through mediators like counterfactual thinking. Counterfactuals are studies have also underlined the negative association between age and procras-
thoughts about what things might have been, particularly focusing on alterna- tination [FBN+ 17]. Further evidence is needed but it may be that the relation
tives to past events; better alternatives are termed upward counterfactuals, while between anxiety and procrastination similarly depends on age and is stronger
worse alternatives are termed downward counterfactuals [ER08] [MGSM93]. among younger people who procrastinate more. Being constantly immersed in
[M.04] suggested that procrastination is overall associated with avoiding up- intense academic settings, those young procrastinators oftentimes do not have
ward counterfactuals (i.e., thinking about how things could have been better) better ways of regulating anxiety and coping with stress; also, both their in-
and making more downward counterfactuals (i.e., thinking about how things accurate metacognitive beliefs about procrastination or themselves and their
could have been worse). When facing an anxiety-provoking task, procrastina- relatively immature executive function system may also contribute to such in-
tors tend to avoid upward counterfactuals and think about how things may have tentional or unintentional avoidance behavior. The conclusion is still speculative
been worse, in response to the anxiety and to restore mood. More importantly, and research is needed to understand whether the relation between anxiety and
Sirois highlighted the crucial role of “the involvement of a self-enhancement procrastination is moderated by age.
motive,” particularly mood repair, in explaining procrastination and procrasti- In addition, findings of the role of gender in the relation between anxi-
nators’ downward counterfactuals. [SP13] made a similar point asserting that ety and procrastination are mixed. According to Beutel and colleagues (2016)
procrastination is very closely related to short-term mood repair and emotion [BMA+ 16], male sex is found to be a predictor of procrastination. However, a
regulation. The findings are again consistent with how the two-factor model considerable amount of studies has found that women are prone to experience
has conceptualized procrastination, in which procrastination functions to re- more anxiety in general compared to males [PMM+ 17]. Such inconsistencies
duce anxiety and repair mood. Though there is currently a lack of evidence on suggest that there may be multiple pathways to procrastination other than
119 120
anxiety. It can be speculated that for women, excessive anxiety may be one of not bidirectional. It is possible that changes in affect following procrastination
the few drivers of procrastination behavior; however, for men, they could end up may take more than one day to occur and that the reverse causal impact of
procrastinating due to a variety of reasons, like poor executive function and bad procrastination to anxiety may be a relatively long-term effect. Future research
time management skills, other than anxiety. Future studies may try to explain may look at the relationship in more varied time frames to determine whether
why such gender difference occurs with more theoretical and empirical support. the result of the study is only limited to a day-to-day relationship between the
variables and whether procrastination increases anxiety.
2.3 Evidence of Procrastination Leading to Anxiety
2.3.1 Evidence of Procrastination Predicting Anxiety Symptoms in
As for the opposite direction of the causal relationship, there is also evidence Psychopathology
suggesting that procrastination can lead to anxiety. [LFF19] noted that teach-
ers reported experiencing significant negative emotions when procrastinating To specify the clinical value of an association between anxiety and procrastina-
and perceiving their dilatory behavior as stressful, pointing out that procras- tion, it is proposed that procrastination may play a key role in predicting psy-
tination may be one of the common stressors that cause anxiety in teachers’ chopathology. [FBN+ 17] examined whether unintentional procrastination can
lives. Such findings may have larger generalizability beyond the domain of the be a marker of common mental disorders using the results of PHQ-9, which
teaching profession and imply the possibility of procrastination being a gen- is used to assess depressive symptoms, and GAD-7, which is used to measure
eral causal factor of anxiety. Furthermore, it is highlighted that Unintentional anxiety symptoms. They pointed out that UPS scores, which measure the level
Procrastination Scale (UPS) scores were independent predictors of anxiety and of unintentional procrastination, are a strong predictor of psychological con-
depression (CITATION). Being a particularly strong marker of anxiety, unin- ditions like depression and excessive anxiety. This implies possible relations
tentional procrastination may indeed cause anxiety and its related symptoms. between mental disorders and procrastination, especially unintentional procras-
In addition, [lRP19] specifically investigated media procrastination, a form of tination, as well as the clinical value of characterizing the role of procrastination
procrastination that involves delaying tasks due to maladaptive engagement in in psychological treatments. Further, it is demonstrated that procrastination is
digital media use. They argued that such procrastination is a key contributor closely correlated with a variety of anxiety-related symptoms, including panic
to negative affect, including stress and anxiety. To better understand the re- disorder, social anxiety disorder, and health anxiety [HPC18]. They also found
sults, they explained that the involvement of off-task media use (OTMU) when that panic disorder symptoms in particular predicted procrastination. Those
facing certain academic tasks can be seen as media procrastination and is of- results are consistent with the assertion that anxiety and procrastination are
ten in conflict with the achievement of long-term academic goals; such ongoing closely associated and underline the potential clinical value of understanding
goal-conflict experience therefore leads to feelings like anxiety, which then per- procrastination and its relation with anxiety.
petuates this cycle of self-regulation failure. This finding is consistent with
the conceptualization of procrastination as a failure of self-regulation whereby
students fail to effectively regulate their use of digital media deviates from aca- 3 Conclusions
demic work, resulting in negative affect. Overall, it is reasonable to speculate Based on the pattern of results reviewed above, it can be concluded that there
that procrastination may lead to anxiety, although more studies that are done is a strong correlation between anxiety and procrastination. Such a conclusion
under different settings and among different samples are needed. is consistent with several widely accepted conceptualizations of procrastination.
Further, some pilot studies have proposed possible cognitive mechanisms be- The correlation is evident among samples of people of different ages, sexes, eth-
tween anxiety and procrastination. [J.09] showed that scores on the Criticism nicities, nationalities, social identities, and living environments. Although most
of Self and Behavior and Difficulty in Achievement questionnaire mediated the of the findings focused on investigating anxiety and procrastination in the aca-
influence of trait procrastination on anxiety, confirming a causal effect of pro- demic domain, the correlation is found in various settings, including academic
crastination on anxiety. These results provide preliminary evidence of a causal and workplace performance, social interacting, bedtime decisions and teams
relationship where procrastination leads to anxiety.
with multiple members. In addition, there is recent neurobiological evidence
However, findings on whether procrastination leads to anxiety are currently
that confirms such correlation and tries to explain it from an anatomical per-
somewhat mixed. In particular, Pollack and Herres (2020) [PH20] conducted a
spective [ZCX+ 20]. Further, some researchers have found a more complex rela-
longitudinal study that measures the level of procrastination and daily negative
tion between anxiety and procrastination and have highlighted the importance
affect (NA) for two weeks and examined whether procrastination leads to a
of some psychological factors like types of anxiety and some external factors
higher level of NA in the following day, controlling for prior levels of affect
like timing and task requirements in understanding the association. Although
(both PA and NA). They found that procrastination did not predict changes in
more r esearch t hat examines different t ypes of procrastination and anxiety un-
NA and highlighted that the relationship between anxiety and procrastination is
121 122
der more diverse social settings is needed to explain the nuanced inconsistencies, expectation of success and the value one places on the goal; however, in the case
overall, the assertion that the correlation between anxiety and procrastination of procrastinators, they are found to be less motivated to initiate tasks that are
is well-supported. more valuable. It suggests the possibility that there may be other cognitive
To specify the direction of this possible causal relationship, it is suggested and emotional factors that also influence people’s, especially procrastinators’,
that anxiety can be a driver of procrastination. There are limited but well- motivation. Further, the time perspective theory fails to explain why engaging
designed studies that show a direct causal effect of anxiety on the development in future time perspective moderates the relation between anxiety and procras-
of procrastination. Moreover, most of the studies have demonstrated anxiety tination; a negative future perspective may as well lead to excessive worries
as a mediator of the relationship between procrastination and other psycholog- about possible consequences of the present behavior and about uncertain future
ical factors, such as resilience, time perspective and some metacognitive beliefs. events, resulting in anxiety. Overall, the conceptualization of procrastination
Overall, it is found that anxiety plays a critical role in leading to the ultimate as an avoidance behavior is highly consistent with the conclusion that anxiety
decision of delaying the initiation or completion of a task. Further, the relation and procrastination are closely associated; future research is needed to examine
between anxiety and procrastination may be mediated by factors like counterfac- how procrastination could fit in the framework of expectancy-value theory and
tual thinking and moderated by factors like gender and age. However, evidence time perspective theory.
for potential mediators and moderators in the relationship between anxiety and
procrastination is limited and more research is needed.
Overall evidence of a causal impact of procrastination on anxiety is inade- 4 Future Directions
quate and mixed, although some researchers have come to the conclusion that
procrastination leads to anxiety. Some studies that implied a possible causal Much of the research on procrastination and anxiety relies on simple correlation
effect of procrastination on anxiety, and others have investigated and confirmed analyses. However, theory and empirical results suggest that procrastination is
the effect in a limited setting or sample. Theoretically, as a self-regulatory a complex phenomenon to which there are multiple psychological pathways. For
example, there is shown to be a strong association between anxiety and procras-
failure, procrastination may lead to anxiety through causing goal-conflict ex-
tination, but there is also an association between depression and procrastina-
periences; empirically, it is noted that certain groups of procrastinators have
reported that their behaviors are debilitating stressful and mediators like Criti- tion [BMA+ 16] [J.09]. We currently do not know whether these emotional fac-
tors have an overlapping connection to procrastination or whether there might
cism of Self and Behavior and Difficulty in Achievement are proposed. Although
be unique relationships between each type of symptoms and procrastination
the evidence is preliminary, it is possible that procrastination leads to anxiety,
behavior. As a first step, researchers should assess multiple relevant factors
forming a vicious cycle.
and include them in more sophisticated models. Examination of the association
Although the field has great clinical potential, findings on the role of pro-
between anxiety and procrastination is a promising and burgeoning field. In
crastination in predicting anxiety disorders are highly limited. Studies have
general, researchers should pay more attention to the influence of anxiety when
suggested that procrastination is closely associated with a variety of mental
looking at procrastination. Instead of only investigating the pure correlation
health problems like depression, social anxiety disorder, and panic disorder.
between anxiety and procrastination, future research should also pay attention
Specifically, some find procrastination to be a strong predictor of depression
to potential causal relationships, whether procrastination leads to anxiety or
and anxiety symptoms. This is consistent with the conclusion that anxiety
whether anxiety leads to procrastination. Clarifying the nature of the relation
and procrastination are closely and strongly associated. Current research has
would be very helpful in understanding procrastination, having large clinical and
agreed on the close relation between procrastination and psychopathology, but
social implications. Particularly, it is important to specify the direction of the
few have thoroughly examined whether and why procrastination is a marker of
causal relationship and to determine whether the causal effect is bidirectional
other clinical symptoms. Further research is needed to verify this prediction role
or unidirectional; identifying the fundamental cause would provide valuable in-
of procrastination and to better understand the clinical value of procrastination
sights into future treatments of both procrastination and anxiety. Specifically,
and its relation with anxiety and psychopathology.
longitudinal studies that are done under varied time frames and regression anal-
The results of this paper are largely consistent with most of the models
ysis that takes more relevant variables like depression and stress into account
about anxiety and procrastination yet have some nuanced inconsistencies with
are needed to better identify a causal influence and to define the correlation.
the others. As explained above, the AAA model and the two-factor model are
Building on the current basis that anxiety and procrastination are strongly asso-
both helpful for understanding the correlation between anxiety and procrastina-
ciated, future research should try to better establish and characterize the causal
tion; they highlight the mechanism of negative reinforcement in the process of
relation.
procrastination being in line with the empirical results and the author’s specula-
Research is also needed to characterize the underlying mechanisms of the
tions. However, when it comes to the expectancy-value theory, a few confusions
relationship between anxiety and procrastination. Knowing why anxiety and
arise. As traditionally viewed, the two general sources of motivation are one’s
123 124
procrastination are so closely correlated and through what processes do they in- [ALEL+ 18] J. Alonso, Z. Liu, S. Evans-Lacko, E. Sadikova, N. Sampson,
fluence one another would help conceptualize procrastination and identify other S. Chatterji, J. Abdulmalik, S. Aguilar-Gaxiola, A. Al-Hamzawi,
relevant psychological factors. Moreover, treatments can be developed to target L. Andrade, R. Bruffaerts, G. Cardoso, A. Cı́a, S. Florescu,
those mediating variables. Future studies may try to conduct mediation analy- G. de Girolamo, O. Gureje, J. Haro, Y. He, P. de Jonge,
sis on potential variables like perfectionism, self-evaluation, self-compassion and E. Karam, N. Kawakami, V. Kovess-Masfety, S. Lee, D. Levin-
level of rumination and worries. Further, more pieces of neurobiological evidence son, M. E. Medina-Mora, F. Navarro-Mateu, Beth-Ellen Pennell,
are necessary to help explain the biological mechanism of the correlation and to M. Piazza, J. Posada-Villa, M. Ten Have, Z. Zarkov, R. Kessler,
solidify or falsify the existing theories. This research area is still young and fresh and G. Thornicroft. Treatment gap for anxiety disorders is global:
as more evidence is required to make stronger and more compelling conclusions. Results of the world mental health surveys in 21 countries. De-
Specifying the intermediate processes between anxiety and procrastination is pression and Anxiety, 2018.
the next step of understanding their relation.
Investigating moderators of the relation is another critical future direc- [AoAAnd] Anxiety and Depression Association of America (ADAA). Un-
tion. Possible moderators include age, gender, culture, socio-economic status derstand the facts: Stress. n.d.
and other cognitive factors like self-esteem and self-efficacy beliefs. Examin- [ASWC18] E. Astrom, A. Seif, B. Wiberg, and M. G. Carelli. Getting
ing whether those factors are moderators helps decide whether the relationship ”stuck” in the future or the past: Relationships between dimen-
exists across all populations or whether certain populations may experience it sions of time perspective, executive functions, and repetitive neg-
differently. Since most of the studies are done under the academic domain, it ative thinking in anxiety. Psychopathology, 2018.
is also important to determine whether an academic or non-academic setting is
a moderator of the relation; if so, this would limit the generalizability of the [B.13] Klingsieck K. B. Procrastination: when good things don’t come
conclusions to students only. Moderators are important determinants of the to those who wait. European Psychologist, 2013.
applicability of all of the above findings. Further research is burdened to prove
the generalizability of such a correlation between anxiety and procrastination. [BD07] M. Balkis and E. Duru. The evaluation of the major characteris-
tics and aspects of the procrastination in the framework of psy-
chological counseling and guidance. Educational Sciences: The-
5 Possible Implications ory and Practice, 2007.
Elucidating the relation between anxiety and procrastination has significant [BGP17] Katharina Böhme, T. Goetz, and Franzis Preckel. Is it good to
implications. Noting that anxiety may be the cause as well as the result of pro- value math? investigating mothers’ impact on their children’s
crastination, it helps people understand some of the psychological mechanisms test anxiety based on control-value theory. Contemporary Edu-
of both anxiety and procrastination. For procrastinators, this could help them cational Psychology, 2017.
develop insights into the emotional factors that have driven their maladaptive
[BHT94] R. Baumeister, T. Heatherton, and D. Tice. Losing control: How
behaviors and how they may get rid of procrastination by solving certain emo-
and why people fail at self-regulation. 1994.
tional problems. For clinicians, such new finding may shift their clinical focus
from being entirely behaviorally-based to paying more attention to the emo- [BM15] B. Bandelow and S. Michaelis. Epidemiology of anxiety disorders
tional and cognitive components of the dilatory behaviors. Treatments that in the 21st century. Dialogues in Clinical Neuroscience, 2015.
target anxiety or stressors or therapies that aim to change maladaptive cog-
nitions may be helpful to solve procrastination. Thus, the conclusions of this [BMA+ 16] Klein Beutel, M. E., E. M., S. Aufenanger, E. Brähler, M. Dreier,
paper may be beneficial for researchers, clinicians, and patients. K. W. Müller, O. Quiring, L. Reinecke, G. Schmutzer, B. Stark,
and K. Wölfling. Procrastination, distress and life satisfaction
across the age range - a german representative community study.
References PloS One, 2016.
[AA15] I. S. Abbasi and R. G. Alghamdi. The prevalence, predictors, [BP00] Allan Blunt and T. A. Pychyl. Task aversiveness and procrasti-
causes, treatment, and implications of procrastination behaviors nation: a multi-dimensional approach to task aversiveness across
in general, academic, and work settings. International Journal of stages of personal projects. Personality and Individual Differ-
Psychological Studies, 2015. ences, 2000.
125 126
[BPCKB18] O. Babaev, C. Piletti Chatain, and D. Krueger-Burg. Inhibition [Ecc83] J. Eccles. Expectancies, values and academic behaviors. 1983.
in the amygdala anxiety circuitry. Experimental & Molecular
Medicine, 2018. [EK77] A. Ellis and W. J. Knaus. Overcoming procrastination. New
York: Signet Books, 1977.
[BRM88] G. Beswick, E. Rothblum, and L. Mann. Psychological an-
tecedents to student procrastination. Australian Psychologist, [ER08] K. Epstude and N. J. Roese. The functional theory of counter-
1988. factual thinking. Personality and Social Psychology Review: An
Official Journal of the Society for Personality and Social Psy-
[BSPT20] E. Bielinis, Jenni Simkin, Pasi Puttonen, and Liisa Tyrväinen. Ef- chology, Inc, 2008.
fect of viewing video representation of the urban environment and
forest environment on mood and level of procrastination. Inter- [Eva90] D. Evans. Problems in the decision making process: A review.
national Journal of Environmental Research and Public Health, Intensive Care Nursing, 1990.
2020. [FBN+ 15] B. A. Fernie, Z. Bharucha, Nikčević, A. V., and M. M. Spada. The
[CAS20] S. J. Chung, H. An, and S. Suh. What do people do before contribution of metacognitions and attentional control to deci-
going to bed? a study of bedtime procrastination using time use sional procrastination. Journal of Rational-Emotive & Cognitive-
surveys. NAME OF JOURNAL, 2020. Behavior Therapy, 2015.
[CBM04] R. Carden, C. Bryant, and R. Moss. Locus of control, test anx- [FBN+ 16] B. A. Fernie, Z. Bharucha, Nikčević, A. V., and M. M. Spada. The
iety, academic procrastination, and achievement among college unintentional procrastination scale. Journal of Rational-Emotive
students. Psychological Reports, 2004. and Cognitive-Behavior Therapy: RET, 2016.
[CDMM05] B. Caldwell, M. Doyle, M. Morris, and Teresa McQuaide. Pres- [FBN+ 17] B. A. Fernie, Z. Bharucha, A. V. Nikčević, C. Marino, and M. M.
encing: Channeling therapeutic effectiveness with the mentally ill Spada. A metacognitive model of procrastination. Journal of
in a state psychiatric hospital. Issues in Mental Health Nursing, Affective Disorders, 2017.
2005. [FDM14] J. R. Ferrari and J. F. Dı́az-Morales. Procrastination and mental
[cdnd] In Merriam-Webster’s collegiate dictionary. Procrastination. n.d. health coping: A brief report related to students. Individual
Differences Research, 2014.
[Che14] S. H. Chen. A survey of resilience and social anxiety on senior pri-
mary school pupils in keelung city (unpublished doctoral disser- [FE95] J. R. Ferrari and R. A. Emmons. Methods of procrastination
tation). national taiwan ocean university, keelung, taiwan. 2014. and their relation to self-control and self-reinforcement: An ex-
ploratory study. Journal of Social Behavior and Personality,
[Cus18] N. Custer. Test anxiety and academic procrastination among 1995.
prelicensure nursing students. Nursing Education Perspectives,
2018. [Fer91] J. Ferrari. Procrastination and project creation: Choosing easy,
nondiagnostic items to avoid self-relevant information. Journal
[DGL01] M. J. Dugas, P. Gosselin, and R. Ladouceur. Intolerance of un- of Social Behavior and Personality, 1991.
certainty and worry: Investigating narrow specificity in a non-
clinical sample. Cognitive Therapy and Research, 2001. [Fer1a] J. R. Ferrari. Compulsive procrastination: Some self-reported
characteristics. Psychological Report, 1991a.
[dJEW02] d J. Eccles and Allan Wigfield. The development of competence
beliefs, expectancies for success, and achievement values from [Fer1b] J. R. Ferrari. A preference for a favorable public impression by
childhood through adolescence. 2002. procrastinators: Selecting among cognitive and social tasks. Per-
sonality and Individual Differences, 1991b.
[DPMM+ 17] V. De Palo, L. Monacis, S. Miceli, M. Sinatra, and S. Di Nuovo.
Decisional procrastination in academic settings: The role of [FSH+ 12] G. Flett, M. Stainton, P. Hewitt, S. Sherry, and C. H. Lay. Pro-
metacognitions and learning strategies. Frontiers in psychology, crastination automatic thoughts as a personality construct: An
2017. analysis of the procrastinatory cognitions inventory. Journal of
Rational-Emotive Cognitive-Behavior Therapy, 2012.
127 128
[FT00] J. R. Ferrari and D. M. Tice. Procrastination as a self-handicap [KSD18] Jana Kühnel, Christine J Syrek, and Anne Dreher. Why don’t
for men and women: A task-avoidance strategy in a laboratory you go to bed on time? a daily diary study on the relationships
setting. Journal of Research in Personality, 2000. between chronotype, self-control resources and the phenomenon
of bedtime procrastination. Frontiers in Psychology, 2018.
[GH19] M. Goroshit and M. Hen. Academic procrastination and aca-
demic performance: Do learning disabilities matter? Current [LFF19] S. Laybourn, A. C. Frenzel, and T. Fenzl. Teacher procrasti-
Psychology, 2019. nation, emotions, and stress: A qualitative study. Frontiers in
psychology, 2019.
[GMC01] K. Gana, B. Martin, and M. D. Canouet. Worry and anxiety: Is
there a causal relationship? Psychopathology, 2001. [LLSN18] A. Legood, A. Lee, G. Schwarz, and A. Newman. From self-
[GO15] Debra M Glick and S. Orsillo. An investigation of the efficacy defeating to other defeating: Examining the effects of leader pro-
of acceptance-based behavioral therapy for academic procrasti- crastination on follower work outcome. Journal of occupational
nation. Journal of experimental psychology. General, 2015. and organizational psychology, 2018.
[HMS98] L. A. Haycock, P. McCarthy, and C. L. Skay. Procrastination in [lRP19] D. B. le Roux and D. A. Parry. Off-task media use in academic
college students: The role of self-efficacy and anxiety. Journal of settings: Cycles of self-regulation failure. Journal of American
Counseling and Development, 1998. College Health, 2019.
[HPC18] T. Hutchison, A. Penney, and J. Crompton. Procrastination and [M.04] Sirois F. M. Procrastination and counterfactual thinking: Avoid-
anxiety: Exploring the contributions of multiple anxiety-related ing what might have been. The British Journal of Social Psy-
disorders. Current Issues in Personality Psychology, 2018. chology, 2004.
[HWPB06] A. J. Howell, D. C. Watson, R. A. Powell, and K. Buro. Aca- [M.14] Sirois F. M. Out of sight, out of time? a meta-analytic investiga-
demic procrastination: The pattern and correlates of behavioral tion of procrastination and time perspective. European Journal
postponement. Personality and Individual Difference, 2006. of Personality, 2014.
[J.09] Hayashi J. [relationship between cognitive content and emotions [MGSM93] K.D. Markman, I. Gavanski, S.J. Sherman, and M.N. McMullen.
following dilatory behavior: Considering the level of trait procras- The mental simulation of better and worse possible worlds. Jour-
tination]. Shinrigaku Kenkyu : The Japanese Journal of Psychol- nal of Experimental Social Psychology, 1993.
ogy, 2009.
[Mil08] C. W. Miller. Procrastination and attention deficit hyperactivity
[KAEB19] Ann-Marie Küchler, P. Albus, D. Ebert, and H. Baumeister. Ef- disorder in the college setting: The relationship between procras-
fectiveness of an internet-based intervention for procrastination tination and anxiety. dissertation abstracts international: Section
in college students (studicare procrastination): Study protocol of b. The Sciences and Engineering, 2008.
a randomized controlled trial. Internet Interventions, 2019.
[Mow47] O. H. Mowrer. On the dual nature of learning—a reinterpreta-
[KBM07] J. D. Kassel, M. Bornovalova, and N. Mehta. Generalized ex- tion of conditioning and problem solving. Harvard Educational
pectancies for negative mood regulation predict change in anx- Review, 1947.
iety and depression among college students. Behavior Research
and Therapy, 2007. [MPT18] U. Metin, M. Peeters, and T. Taris. Correlates of procrastination
and performance at work: The role of having “good fit”. Journal
[KC19] C. A. Ko and Y. Chang. Investigating the relationships among re- of Prevention Intervention in the Community, 2018.
silience, social anxiety, and procrastination in a sample of college
students. Psychological Reports, 2019. [MT99] N. Milgram and Y.E Toubiana. Academic anxiety, academic pro-
crastination, and parental involvement in students and their par-
[KLMSD+ 19] H. Kaya Lefèvre, C. Mirabel-Sarron, A. Docteur, V. Leclerc, ents. British Journal of Educational PsychologyL, 1999.
A. Laszcz, P. Gorwood, and C. Bungener. Time perspective dif-
ferences between depressed patients and non-depressed partici- [MWB91] A. MacLeod, J. Williams, and D. Bekerian. Worry is reason-
pants, and their relationships with depressive and anxiety symp- able: the role of explanations in pessimism about future personal
toms. Journal of Affective Disorders, 2019. events. Journal of abnormal psychology, 1991.
129 130
[NL20] A. Ng and P. F. Lovibond. Self-efficacy moderates the relation- [RNS+ 08] P. Rosário, J. C. Núñez, A. Salgado, J. A. González-Pienda,
ship between avoidance intentions and anxiety. Emotion (Wash- A. Valle, C. Joly, and A. Bernardo. [test anxiety: Associations
ington, D.C.), 2020. with personal and family variables]. Psicothema, 2008.
[NLL11] Y. Nie, S. Lau, and Albert K. Liau. Role of academic self-efficacy [RRBvdL18] Marie My Lien Rebetez, L. Rochat, C. Barsics, and M. van der
in moderating the relation between task importance and test anx- Linden. Procrastination as a self-regulation failure: The role of
iety. Learning and Individual Differences, 2011. impulsivity and intrusive thoughts. Psychological Reports, 2018.
[OC01] A. Onwuegbuzie and K. M. Collins. Writing apprehension and [RSM86] E. D. Rothblum, L. J. Solomon, and J. Murakami. Affective,
academic procrastination among graduate students. Perceptual cognitive, and behavioural differences between high and low pro-
and Motor Skills, 2001. crastinators. Journal of Counseling Psychology, 1986.
[PAZ18] S. Pearlman-Avnion and A. Zibenberg. Prediction and job- [SBE11] Laura C. Selkirk, H. Bouchey, and J. Eccles. Interactions among
related outcomes of procrastination in the workplace. Journal domain-specific expectancies, values, and gender: Predictors of
of Prevention & Intervention in the Community, 2018. test anxiety during early adolescence. The Journal of Early Ado-
lescence, 2011.
[PH20] S. Pollack and J. Herres. Prior day negative effect influences cur-
rent day procrastination: A lagged daily diary analysis. Anxiety, [Sch07] Harvard Medical School. National comorbidity survey (ncs).
Stress, & Coping, 2020. (2017, august 21). data table 2: 12-month prevalence dsm-
iv/wmh-cidi disorders by sex and cohort. 2007.
[PMAP00] V. Prohaska, P. Morrill, I. Atiles, and A. Perez. Academic pro-
[SF13] P. Steel and J. Ferrari. Sex, education and procrastination: An
crastination by nontraditional students. Journal of Social Behav-
epidemiological study of procrastinators’ characteristics from a
ior and Personality, 2000.
global sample. European Journal of Personality, 2013.
[PMM+ 17] M. Paechter, D. Macher, K. Martskvishvili, S. Wimmer, and
[SFP+ 18] C. Schlüter, Christoph Fraenz, M. Pinnow, P. Friedrich,
I. Papousek. Mathematics anxiety and statistics anxiety. shared
O. Güntürkün, and E. Genç. The structural and functional sig-
but also unshared components and antagonistic contributions to
nature of action control. Psychological Science, 2018.
performance in statistics. Frontiers in Psychology, 2017.
[Sir07] M. Sirois, F. “i’ll look after my health, later”: A replication and
[PtDA15] H. L. Phillips, T. 4th, Dong, S. J. Durning, and Jr E Artino,
extension of the procrastination-health model with community-
A. R. Assessing task importance and anxiety in medical school:
dwelling adults. Personality and Individual Differences, 2007.
An instrument development and initial validation study. Military
Medicine, 2015. [SJ01] J. Stoeber and J. Joormann. Worry, procrastination, and per-
fectionism: Differentiating amount of worry, pathological worry,
[RBF+ 18] A. Rozental, S. Bennett, D. Forsström, D. D. Ebert, R. Shafran, anxiety, and depression. Cognitive Therapy and Research, 2001.
G. Andersson, and P. Carlbring. Targeting procrastination us-
ing psychological treatments: A systematic review and meta- [SJ19] J. Song and Y. Jiang. The distinct roles of proximal and distal
analysis. Frontiers in Psychology, 2018. utility values in academic behaviors: Future time perspective as
a moderator. Frontiers in Psychology, 2019.
[RF20] Catherine A. Roster and J. Ferrari. Time is on my side—or is
it? assessing how perceived control of time and procrastination [SLF00] M. Stainton, C. H. Lay, and G. L. Flett. Trait procrastinators
influence emotional exhaustion on the job. Behavioral Sciences, and behavior/trait-specific cognitions. Journal of Social Behavior
2020. and Personality, 2000.
[RFNU11] L. A. Rabin, J. Fogel, and K. E. Nutter-Upham. Academic pro- [SM13] M. Skowronski and A. Mirowska. A manager’s guide to workplace
crastination in college students: The role of self-reported execu- procrastination. SAM Advanced Management Journal, 2013.
tive function. Journal of Clinical & Experimental Neuropsychol- [SP13] F. Sirois and T. Pychyl. Procrastination and the priority of short-
ogy, 2011. term mood regulation: Consequences for future self. Social and
Personality Psychology Compass, 2013.
131 132
[SR84] L. J. Solomon and E. D. Rothblum. Academic procrastination: [ZH18] S. Zacks and M. Hen. Academic interventions for academic pro-
Frequency and cognitive-behavioral correlates. Journal of Coun- crastination: A review of the literature. Journal of Prevention
seling Psychology, 1984. Intervention in the Community, 2018.
[SWS00] L. J. Schubert Walker and D. W. Stewart. Overcoming the pow- [ZLF19] S. Zhang, P. Liu, and T. Feng. To do it now or later: The cog-
erlessness of procrastination. Guidance & Counseling, 2000. nitive mechanisms and neural substrates underlying procrastina-
tion. Wiley Interdisciplinary Reviews. Cognitive Science, 2019.
[TB97] D. M. Tice and R. F. Baumeister. Longitudinal study of procras-
tination, performance, stress, and health: The costs and benefits
of dawdling. Psychological Science, 1997.
[TS89] B. Tuckman and T. Sexton. Effects of relative feedback in over-

coming procrastination on academic tasks. paper presented at the
meeting of the american psychological association, new orleans,
la. 1989.
[vE03] W. van Eerde. A meta-analytically derived nomological network

of procrastination. Personality and Individual Differences, 2003.
[VFGI12] S. Vahedi, F. Farrokhi, F. Gahramani, and A. Issazadegan.

The relationship between procrastination, learning strategies and
statistics anxiety among iranian college students: a canonical cor-
relation analysis. Iranian Journal of Psychiatry and Behavioral
Science, 2012.
[vHVM18] E. van Hooft and H. Van Mierlo. When teams fail to self-regulate:
Predictors and outcomes of team procrastination among debating
team. Frontiers in Psychology, 2018.
[VW09] S. P. Vassilopoulos and E. R. Watkins. Adaptive and maladaptive
self-focus: A pilot extension study with individuals high and low
in fear of negative evaluation. Behavior Therapy, 2009.
[WM96] A. Wells and G. Matthews. Modelling cognition in emotional
disorder: The s-ref model. Behavior Research and Therapy, 1996.
[YWH+ 17] Yi-Chun Yeh, Peng-Wei Wang, Mei-Feng Huang, Pai-Cheng Lin,
and C. Ko. The procrastination of internet gaming disorder in
young adults: The clinical severity. Psychiatry Research, 2017.
[ZCX+ 20] R. Zhang, Z. Chen, T. Xu, L. Zhang, and T. Feng. The overlap-
ping region in right hippocampus accounting for the link between
trait anxiety and procrastination. Neuropsychologia, 2020.
[ZDF+ 18] Yanting Zhang, Siqin Dong, W. Fang, Xiaohui Chai, Jiaojiao
Mei, and Xiuzhen Fan. Self-efficacy for self-regulation and fear of
failure as mediators between self-esteem and academic procrasti-
nation among undergraduates in health professions. Advances in
Health Sciences Education, 2018.
133 134
1 Purpose:
In finding the most pathogenic mutation of the known missense mutations, the
goal is to raise awareness for those with specific mutations to be more careful
Prioritizing missense mutations in BDNF to so they can decrease their chances of developing AD or AD-like symptoms. The
BDNF Val66Met polymorphism should be considered as a target for the novel
predict variant pathogenicity in Alzheimer’s Alzheimer’s disease therapeutics.
Disease (AD)
2 Hypothesis:
∗
Srihas Rao If the val66met polymorphism occurs in the BDNF protein, then the protein
will be most likely to cause Alzheimer’s related symptoms among the mutations
April 2, 2021 that we studied. We predict this because there have been previous studies that
show that the V66M mutation causes reduced BDNF protein in the brain, thus
causing Alzheimer’s and Alzheimer’s related symptoms.
Abstract
The number of elder people will double from 2000 to 2050. When older, 3 Introduction
people become more susceptible to neurodegenerative disorders. One gene
that affects the phenotypes of neurodegenerative diseases is Brain-Derived What is BDNF?
Neurotrophic Factor, BDNF. BDNF is a protein with 5 different isoforms Brain-derived neurotrophic factor (BDNF), a molecule known to regulate
in the human Chromosome number 11. For this study, only missense mu- neuronal survival and plasticity, is widely expressed in the developing and adult
tations were analyzed. By limiting the analysis, we can develop strategies mammalian brain [ZYC08]. Alzheimer’s is an illness that is a dynamic, unal-
to predict potential pathogenic effects. These missense mutations could terable mind issue that gradually demolishes memory and thinking aptitudes
be one of the risk factors for developing neurodegenerative diseases. The
and, inevitably, the capacity to complete the most straightforward errands. In
bioinformatics tools MARRVEL, NCBI, Clustal Omega, STRING, and
the vast majority with the sickness, side effects initially show up in their mid-
TMHMM 2.0 were used to analyze the mutations. We extracted missense
mutations data, related parameters through Geno2MP, and added amino 60s [Fac19]. The disease is multifactorial, meaning that the disease mechanism
acid change, conserved up to, and an amino acid change position in the has both genetics and epigenetics contribution [BDN20]. People’s lifestyle, diet,
domain columns. In addition to this analysis, the structural analysis of and environment are also involved [Fac19]. Thus, while there are multiple genes
the well known pathogenic mutations of BDNF were analyzed. With this that can increase the risk to develop this disorder, it does not guarantee that one
data, the most potential damaging mutation was found by prioritizing will get the disorder. Due to the complexity, the disease mechanism is not fully
characteristics of the mutations, and we determined it is V66M. understood, but one of the potential candidates to understand it is BDNF, or
Brain Derived Neurotrophic Factor. It is a potential factor because there have
Keywords: Alzheimer’s Disease, BDNF, missense, pathogenicity, Geno2MP,
been many studies [JB15] showing that the BDNF gene has caused an increased
MARRVEL
risk of older people developing some of the symptoms like those of Alzheimer’s.
Effect of BDNF on Humans
The reduction of BDNF has shown to cause problems in elderly people
[ZYC08]. These problems include short term memory loss, difficulty complet-
ing familiar tasks, and confusion with time or place, all common symptoms of
Alzheimer’s Disease. Some interventions like exercise or antidepressant admin-
istration, enhance the expression of BDNF in normal and pathological condi-
tions. Thus, if one stays active, he or she can have a reduced risk of developing
Alzheimer’s [ZYC08].
The BDNF gene encodes a protein called a brain-derived neurotrophic factor,
found in the brain and spinal cord, and localized in the hippocampus [BDN20].
∗ Advised by: Zeynep Öztürk, University Of Cambridge This protein advances the endurance of nerve cells by assuming a job in the
135 136
development. In the brain, the BDNF protein capacities at the associations more deleterious missense mutations. With this information, we can sort and
of neural connections, where cell-to-cell correspondence happens. The neural filter the different characteristics of each mutation to better understand which
connections can change and adjust after some time considering understanding, mutations are more harmful than the others. When filtering, the more important
a trademark called synaptic plasticity [JB15]. characteristics can be prioritized.
Effects of reduced BDNF: To see how this polymorphism and others affect the phenotype of Alzheimer’s
Changes in BDNF articulation are related to both typical and neurotic ma- Disease, we used the bioinformatic tools MARRVEL (Geno2MP) and NCBI
turing and mental sickness, in structures significant for memory procedures, (protein database) to compare different alterations of BDNF gene to prioritize
for example, the hippocampus and parahippocampal regions. BDNF is urgent more deleterious missense mutations. With this information, we can analyze the
to learning and memory since it directs long term depression (LTD) and long- various scores that tell information about the mutations, like Grantham Score,
term potentiation (LTP), synaptic versatility, axonal growing, multiplication of PolyPhen2 Score, and Conserv Score. Grantham score predicts the effect of the
dendritic arbour, and neuronal separation [JB15]. In addition, reduced BDNF polymorphisms based on the chemical properties, like polarity and molecular
messenger RNA and protein levels have been found in the hippocampus and size, and PolyPhen2 Score is the probability that a mutation is harmful. A
other cortical areas in patients with AD. Thus, mutated BDNF proteins can score below 50 for the Grantham score is considered more harmful and a score
lose functionality so there will be lowered protein levels and many long term above .80 for PolyPhen2 is considered pathogenic. Also, Conserv Score tells
ailments can arise. how conserved a mutation is.
Isoforms in BDNF: Predicting effect of BDNF on Alzheimer’s with modeling:
In Brain Derived Neurotrophic Factor, there are five isoforms, with the Additionally, the structural effects of the mutation can be seen by modeling
longest being 247 amino acids long. Isoforms occur when a gene is trascripted the proteins with and without the mutations using UniProt to get the FASTA
from the same locus but are different in their transcription start sites. The sequence and PyMol modeller Server to model them. By looking at these alter-
sequences that are common among the five isoforms are the more important se- ations we tried to understand the causative reasons for the protein dysfunction.
quences because they are used every time. When modeling, we will look at the In this study, we analyzed the structural features of the BDNF and predicted
five isoforms and how they structurally change with the most influential muta- the potential pathogenic or non-pathogenic alleles reported in databases. We
tions found from the bioinformatic analysis. This will help us understand which used bioinformatic tools, such as: TMHMM which a server to predict transmem-
of the mutations found from the bioinformatic analysis are showing significant brane domain in BDNF protein, Clustal Omega which can compare the FASTA
structural problems resulting in AD symptoms. sequences of BDNF from different species, STRING to see proteins interacting
Mutations in BDNF: with BDNF, and Geno2MP in MARRVEL to extract and compare the missense
To predict the effect of BDNF on AD, one approach could be analyzing mutations in BDNF. Once we found what was the most harmful mutation from
missense alterations in the gene. In each human genome, there are polymor- the bioinformatic analysis, we used the Swiss model server in order to model the
phisms, slight changes in genes that result in genetic variation, making each proteins to see the structural changes that occurred because of the mutations.
human genome distinct from others. An example of mutations in BDNF im- In finding the most pathogenic mutation of the known missense mutations, the
pacting Alzheimer’s is the Val66Met polymorphism is implicated in synaptic goal is to raise awareness for those with specific mutations to be more careful
excitation and neuronal integrity, and has previously been shown to moder- so they can decrease their chances of developing AD or AD-like symptoms.
ate amyloid-β-related memory decline and hippocampal atrophy in preclinical
sporadic Alzheimer’s disease. From previous studies, the val66met polymor-
phism has influenced memory in people from ages 20-93 [ZYC08]. In studies of 4 Materials and Methods:
brain morphometry using structural magnetic resonance imaging (MRI) scans,
A Computer with high speed internet access and online Kinematics tools were
Val/Met individuals have repeatedly been shown to have a smaller hippocam-
required.
pal volume relative to controls which are homozygous for Val allele [ZYC08].
In other studies, it is shown that Met66 carriers showed greater dysfunction in
cognition, glucose metabolism and tau, with implications for clinical trial de- 4.1 MARRVEL/Geno2MP
sign [YYL16]. Finally, Val66Met also has shown an increased risk of developing
Individuals with a missense mutation in the BDNF/BDNF-AS gene can be
AD in women, and Caucasian women, specifically [FMN].
found using Geno2MP (2020). With this data, an excel sheet was made and the
Predicting effect of BDNF on Alzheimer’s with Bioinformatics:
columns, amino acid change, hydrophilic/hydrophobic change, significance, and
To see how this polymorphism and others affect the phenotype of Alzheimer’s
conservation were added. Then, with the amino acid change, a hydrophilic/hydrophobic
Disease, we used the bioinformatic tools MARRVEL (Geno2MP) and NCBI
(protein database) t o compare different alterations of BDNF gene t o prioritize
137 138
change can be found by seeing if it changed from a hydrophobic protein to a hy- 4.4 Analyzing secondary structure changes in Polymor-
drophilic protein, vise versa, or stayed the same structure/chemistry/property. phisms
If the property stayed the same, the change is not significant. If the structure
did change radically, then it is significant. - Find Individuals with a missense 1. Using www.Uniprot.org, collect the data for BDNF Natural variant P23560
mutation in the BDNF/BDNF-AS gene using Geno2MP (2020) 2. DOPE scores were found for all of the known missense mutations
- Create an excel sheet with the columns amino acid change, hydrophilic/hydrophobic
change, significance, and conservation 3. The mutations with the least DOPE scores were T2I, V66M, Q75H, M122T,
- With the amino acid change, a hydrophilic/hydrophobic change can be R125M, and R127L.
found by seeing if it changed from a hydrophobic protein to a hydrophilic pro-
tein, vise versa, or stayed the same. 4. Preparing file Multiple Sequence Alignment
- Check the structure; If there is a change then it is significant otherwise it - Copy the FASTA format sequence of all 5 isoforms in text document
is not significant. Check for Grantham score, gerpscore, and PolyPhen2 score. - The fasta format of BDNF gene isoforms paste into the large text box
in SOPMA and submit to get the results.
4.2 Missense mutation positioning in NGF Domain 5. Homology modeling/ tertiary structure prediction
- Use modeller software to modelle Structures
Proteins have domains which are amino acids generating functional regions in
- BLAST was performed and 3QB5 was selected as target
the protein. If there is a mutation in a functional domain, it is expected that this
- Structures were modelled by running Python script
mutation might affect the protein function. To interpret whether the amino acid
- Structures with least DOPE score were selected for superimposing
changes negatively affect the protein function we need to know the positions of
- The reference structure and modelled structure were imported to PyMol
these changes. We used MARRVEL/DIOPT 7.1 interphase to know the NGF
and commands were performed to superimpose the two structure
domain (amino acids 212-329) of the isoform ‘NP001137282.1 [MMB19]. The
data was used from Geno2MP is for the isoform sp|P 23560.1|. NGF domain
was found manually in the isoform sp|P 23560.1| as it is the analyzed isoform in
Geno2MP. NCBI was used to have FASTA amino acid sequences of these two 5 Results
isoforms and the NGF region was detected after having alignment from Clustal
Omega. Amino acids were counted, and changes were checked whether they are 5.1 MARRVEL/Geno2MP
in the NGF domain or close to the domain.
5.1.1 BDNF Missense mutations reported in Geno2MP
- Download FASTA amino acid sequences of the isoform sp|P 23560.1| from There are 25 missense mutations reported. The data comes with many char-
NCBI
acteristics like gene information ( Chr:Pos, Alleles) and protein information (
Protein change, amino acid change, and significance) as well as pathogenicity
- Find NGF domain manually in the isoform sp|P 23560.1| using data col-
prediction. Here we tried to improve that pathogenicity prediction and apply
lected from Geno2MP in step 4.1
an approach that we can filter missense variants more specific.
5.1.2 Prioritizing the most pathogenic variants with 2 different strate-

4.3 Analysing missense alleles for conservation/position gies
To determine the severity of a mutation, criteria such as gerpscore, grantham Strategy 1: In this strategy, we sorted and filtered the table based on gerp-
score, conservation of amino acids and pathogenicity prediction from Geno2MP score biggest to smallest, pathogenicity, important change II.
were used. Using Table 1, the several strategies were applied. For example, for Strategy 2: In this strategy, we sorted and filtered the table based on polyPhen2
strategy 1 we applied sorting/filtering in the order gerp score biggest to smallest,
score → pathogenicity → Grantham score.
pathogenicity, important change II. To apply this strategy, we first sorted the
gerpcores from highest to lowest. Then, Grantham scores are filtered as higher
100. By filtering the important traits by following different strategies, the most 5.2 5 missense alleles were found in NGF domain of BDNF
likely damaging mutation can be found. We tried different strategies to get the As shown in Figure 1, the 2 isoforms have the same domains, but the coordinates
most possible damaging alterations and to test our approaches. are different. From amino acids 83-329 i n t he l ong i soform, t he s equence i s
139 140
repeated in the short isoform. In addition, the NGF domain can be found from 5.4 Secondary and tertiary structural analysis
the amino acids 212-329. This means that 5 of the mutations are in the NGF
The Root Mean Square Deviation (RMSD) of two aligned structures indicates
domain.
their divergence from one another. In Pymol RMSD will be printed as RMS and
the units are Angstroms. Pymol shows the structure changes in the mutated
5.3 Conservation of mutated amino acids in Isoforms protein as cyan and the wild type protein in green. T2I shows no significant
Conservation is an important feature for an amino acid. Throughout evolu- structural changes, and as a RMS score of 0.233.
tion it is observed that essential amino acids are conserved among organisms V66M shows a significant structural change because of how the sticks differ
for the proteins which have similar functions. We checked BDNF protein and where Figure 4 has the protein structure highlighted. Additionally, this muta-
tested conservation of amino acids which carry missense mutations. The Clustal tion also had a RMS score 0.364, the highest among the mutations. Therefore,
Omega comparing the eight FASTA sequences of the isoforms shows which based on our structural analysis, the most pathogenic missense mutation in
amino acids are conserved among different isoforms. The number of isoforms BDNF is the Val66Met mutation.
that are conserved of a specific amino acid location can be used to determine Homology modeling/ tertiary structure prediction
if a mutation is conserved or not. The mutation that occurs at the 66th and
120th position, V66M and A120T, is conserved in isoforms 1, 7, 4, 5, and 6. The
mutations that occur at the 144th position, and I144T. The amino acids that
6 Data Analysis:
are conserved in all species are favored over others when analyzing the data. Alzheimer’s Disease is one of the common neurodegenerative diseases and is
currently the sixth leading cause of death [Fac19]. It affects mostly elderly peo-
ple and makes them dependent on caregivers. There is no cure and the disease
mechanism is not completely understood. BDNF is one of the potential players
in the disease mechanism [Fac19]. BDNF promotes the survival and differen-
tiation of selected neuronal populations of the peripheral and central nervous
systems. Therefore, a decrease in BDNF protein in the brain can result in AD-
like symptoms. There are several missense mutations reported in the Geno2MP
database generated by the human genome project. Missense mutations may
or may not affect protein function. If there is a missense mutation generating
dysfunctional protein this could increase the risk of developing AD or cause
some other detrimental symptoms. Here we showed that bioinformatic analysis
could help us to understand and prioritize some of these mutations for future
research. In doing this research, we have found that based on our analysis,
V66M was the most pathogenic mutation. This was determined by focusing on
the chemical structure and nature of the protein and finding which amino acid
changes were significant, conserved and in the NGF domain, and thus the effect
on the secondary and tertiary structure. We tried different strategies to have
the best potential prediction. With these findings, we predict that those with a
genetic history of these pathogenic mutations could be in the risk of developing
AD or AD-like symptoms. There are no reported cases diagnosed with AD and
carrying those mutations. Our prediction is based on having a dysfunctional
protein which eventually could lead to neurodegeneration and AD mechanism
or similar outcomes. However, we cannot be sure without experiments which
test the effect of the mutation in vivo.
Once we can define the most pathogenic mutations, we will have a better
understanding of the disease mechanism. Since the amino acid changes cause
alterations in the protein’s property, the cause of Alzheimer’s disease can be
found on a molecular level and this could help us understand the heterogeneity
141 142
of the disease. ngf and gdnf in aging and alzheimer’s disease. aging and disease,. Ag-
ing & Disease (A&D), 2015.
7 Conclusion [MMB19] Marı́a Belén Zanoni Magdalena Miranda, Juan Facundo Morici
and Pedro Bekinschtein. Brain-derived neurotrophic factor: A key
AD is one of the mysterious diseases that we need to solve. Finding a cure for molecule for memory in the healthy and the pathological brain. Fron-
the disease is based on the knowledge we have about disease mechanisms. Here tiers, 2019.
we tried to develop a strategy based on chemical changes in BDNF protein due
to missense mutations in the gene. We extracted data from MARRVEL. The [Str20] functional protein association networks. (n.d.). string-db.org.
chromosome position (Chr:Pos), alleles, gene, annotations, protein change, hy- https://string-db.org/cgi/network.pl?taskId=lx4SVFq7KjS4, 2020.
drophilic/hydrophobic change, significance, conservation in other species, NGF
[TMH20] Tmhmm server, v. 2.0. Www.Cbs.Dtu.Dk.
domain, and pathogenicity was found by using the tools Geno2MP/MARRVEL,
http://www.cbs.dtu.dk/services/TMHMM/, 2020.
Clustal Omega, String, ClinVar, and NCBI. After gathering all the data (Table
1), the most pathogenic mutations were found by sorting and filtering for the [YYL16] Carlos Cruchaga Alison Goate Anne M. Fagan Tammie L.S. Ben-
strategies we defined in the Results (3.6). Results showed that like what is stated zinger Paul Maruff Peter J. Snyder Colin L. Masters Ricardo Al-
in the hypothesis, the most pathogenic mutations was V66M based on possibly legri Jasmeer Chhatwal Martin R. Farlow Neill R. Graff-Radford
having a dysfunctional protein which eventually could lead neurodegeneration Christoph Laske Johannes Levin Eric McDade John M. Ringman
and AD mechanism or similar symptoms. However, there are other pathogenic Martin Rossor Stephen Salloway Peter R. Schofield Yen Ying Lim,
mutations and thus we can analyze missense variants for their pathogenicity Jason Hassenstab. Bdnf val66met moderates memory impairment,
to help us understand disease mechanisms. This can help us show what mu- hippocampal function and tau in preclinical autosomal dominant
tations have an effect on Alzheimer’s Disease or AD related symptoms (6, 7). alzheimer’s disease. Mayo Clinic, 2016.
Although having this mutation cannot be changed, individuals can still not de-
velop Alzheimer’s Disease or AD-like symptoms. This could be either due to [ZYC08] Bruce McEwen Barbara Hempstead Francis Lee Zhe-Yu Chen 1,
their environment, diet, life standards, and mental activities or the complexity Kevin Bath. Impact of genetic variant bdnf val66met on brain struc-
of the AD mechanism. We are not claiming that carrying one pathogenic vari- ture and function. PubMed.gov, 2008.
ant could lead AD alone. However, it would be beneficial to consider studying
potential pathogenic variants to understand the disease mechanism.
8 Acknowledgement
In the successful completion of this project, I would like to thank my family for
their help and guidance throughout the project.
References
[BDN20] Bdnf gene: Medlineplus genetics. (n.d.). medlineplus.gov.
https://medlineplus.gov/genetics/gene/bdnf/#normalfunction,
2020.
[Fac19] Alzheimer’s disease fact sheet. NCBI, 2019.
[FMN] Critical issues in bdnf val66met genetic studies of neuropsychiatric

disorders.
[JB15] Francielle Mina Michelle Lima Garcez Alexandra Ioppi Zugno

Josiane Budni*(), Tatiani Bellettini-Santos. The involvement of bdnf,
143 144
Figure 2: Table 2: Grantham Score → Is the change important II. For this
strategy, we started with filtering all Grantham score values higher than 50 out.
Then, we chose only yes in the “is the change still Important?” column. From
this filtering, we obtained these 8 mutations.
Figure 3: PolyPhen2 score → pathogenicity → Grantham score. For this strat-

egy, we started with filtering all PolyPhen2 score values less than 0.8 out. Then,
we filtered the pathogenicity column to only have probably damaging, risk fac-
tor, and unknown. Finally, filtering out all Grantham score values higher than
50 out. From this filtering, we obtained these 11 mutations.
Figure 1: Missense mutations of BDNF and BDNF-AS reported in the

Geno2MP database. 25 missense mutations were found in the Geno2MP
database and analyzed for their effect on protein nature. In this table we have
information about chromosome position (Chr:Pos), Alleles, gene, annotations,
protein change, hydrophilic/hydrophobic change, significance, conservation in
other species, NGF domain, and Pathogenicity.
Figure 4: NGF domain of two BDNF isoforms. NP 001137282.1 is isoform e of

human and sp|P 23560.1| form is used by Geno2MP.
145 146
Figure 6: Sopma results of missense mutation T2I with sequence length 247.
The figure shows the RMS of 0.233 with Superimpose of the two structures in
Pymol and mutated protein as cyan and the wild type protein in green.
Figure 5: Clustal Omega comparison of BDNF in Isoform 1-8. The figure shows
the alignment of 8 different isoforms of the BDNF protein. The figure shows
a symbol, either a star, color, a period, or nothing below. A star signifies an
amino acid that is the same among all of the species, a colon means it is similar
in all but one, and a period means it is random. In addition, when there is a
dash in the sequence, that means that part of the sequence is not there.
Figure 7: Sopma results of missense mutation V66M with sequence length

247. The figure shows RMS of 0.364 with Superimpose of the two structures in
147 148
Figure 8: Sopma results of missense mutation Q75H with sequence length 247.
The figure shows RMS of 0.219 with Superimpose of the two structures in Pymol Figure 10: Sopma results of missense mutation R125M with sequence length
and mutated protein as cyan and the wild type protein in green. 247. The figure shows RMS of 0.199 with Superimpose of the two structures in
Figure 9: Sopma results of missense mutation M122T with sequence length 247.
The figure shows RMS of 0.214 with Superimpose of the two structures in Pymol
and mutated protein as cyan and the wild type protein in green. Figure 11: Sopma results of missense mutation R127L with sequence length
247. The figure shows RMS of 0.219 with Superimpose of the two structures in
149 150
A Perspective on Novel Proteins
∗
Maggie Lau
April 2, 2021
Abstract
Proteins facilitate many necessary processes for life in our bodies and
in nature. Since the late 1900s, scientists have been able to design novel
proteins, which are proteins not found in nature. These novel proteins
have shown to possess more desirable properties than proteins found in
nature, and thus can be used to benefit society. They are usually made
by modelling their structure computationally first, then synthesized using
techniques such as recombinant methods. Two major methods of mod-
elling are template modelling, in which there is a known template such as
the protein sequence, or de novo, in which the protein structure is built
from the bottom up. With modelling, the goal is to find a thermodynam-
ically stable conformation that the protein folds into. However, designing
proteins is not easy, and among many of the challenges is Levinthal’s para-
dox, which states there are too many protein conformations to sample.
Despite these challenges, there are countless areas where novel proteins
can be used, and this paper details some uses of these proteins in the
medical field and industrial field.
Figure 12: Represents RMS score and Mutations with error bars.
1 Introduction
Proteins are natural biological molecules that perform many essential functions,
such as cellular division, in our bodies and in nature. Considering that proteins
are made in nature, it is remarkable that scientists are capable of designing
proteins. Their main goal is to create proteins that have customized functions
and in fact, research into protein design has been done since the 1900s. Among
the first synthetic proteins created was ribonuclease, and this particular protein,
which plays important roles in RNA metabolism, was designed by Ralph F.
Hirschmann and his team [Hev09]. As protein design advanced over the years,
scientists have been able to create novel proteins, which are completely new
proteins that are not found in nature. This paper will discuss the reasons
for making novel proteins, what to consider before making novel proteins, the
methods used to make novel proteins and its challenges, and the applications of
novel proteins.
∗ Advised by: Jacob Kirsch from Stanford University
151 152
2 Novel Proteins and non-template based modelling (de novo), in which the model of the protein
structure is generated from scratch [BBB20].
Novel proteins can either be made from scratch, or based on a similar pro-
tein [n.andb], and to make these novel proteins, scientists first identify the tar- While the approach to modelling proteins can vary, there are some common
get protein structure (their desired protein structure). The protein can then basic steps. For template based modelling, a similar sequence is identified from
be synthesized, for instance, by expanding the genetic code and assigning non- a protein database such as The Protein Data Bank (PDB) [n.ande], and a cri-
canonical amino acids to a stop codon or having a tRNA carry it with an- teria is designed for comparing the similarities between the target sequence and
other amino acid [PDRK12]. In addition, recombinant methods may also used the template sequences. Afterwards, the sequences are aligned to start mod-
and organisms such as Escherichia coli (E. coli ) can be used to mass produce elling. In a type of template based modelling called comparative modelling, or
the proteins. Other approaches to synthesizing proteins include directed evolu- homology modelling, a 3D structure and an atom model of a target protein is
tion [RECZ12], in which the protein is modified to acquire the desired properties. built based on similar sequences [Fis10]. Another method, protein threading,
models structures in which only similar protein folds are known [PX09].
As it turns out, novel proteins have many advantages, such as performing func-
tions as well as or even better than existing proteins. For example, an artificial For cases where a template does not exist, de novo modelling can be used.
enzyme was able to catalyze CO2 hydration with an efficiency comparable to In this type of modelling, a target protein backbone (the primary sequence of
some naturally occurring carbonic anhydrases and within 350-fold of the fastest amino acids) that will fold into the desired structure is identified. Then sim-
isozyme, CAII [DeN]. In addition, scientists have more control when designing ulations are run and it follows different mathematical parameters and finds a
the functions of proteins because the conformations of natural proteins are only thermodynamically favorable structure. After the protein samples many pos-
a subset of the possible conformations that proteins can have. Consequently, sible structural ensembles, the combinations of proteins are usually narrowed
novel proteins can take on different conformations that may allow them to per- down, and the generated protein model is refined to resemble the native struc-
form functions not found in nature [Bak19]. Ultimately, creating novel proteins ture (the structure that the protein naturally folds into). In modelling protein
allows scientists to have a better understanding of how proteins fold and their structures, computational methods such as Monte Carlo Simulation, which uses
functions because they can alter a part of the protein such as its amino acid random sampling to obtain a numerical value, are used very often. Another
sequence and study the effects of the modification [MM05]. widely used method is molecular dynamics, in which Newton’s laws are ap-
plied to track the movement of atoms. In general, these algorithms may be
Even so, there are many things to consider in order to design novel proteins, deterministic, in which the output is determined by the input or parameters,
particularly depending on whether the primary structure or the tertiary struc- or stochastic, in which an answer may be determined randomly [KB19]. Then
ture is used as the starting point when modelling protein structure. If beginning for selecting from these thermodynamically favorable structures and evaluating
with primary structure, one must consider the protein folding problem - can a how close they are to the native structure, force fields such as physics based
protein’s structure be predicted from its amino acid sequence? Scientists also energy fields, which are governed by the laws of physics, and knowledge based
need to identify the target structure that they are aiming to create, or begin energy functions, which are based on information from experimentally solved
to design from a tertiary structure. To do this, they must contend with the structures, are used [JLZ17].
inverse protein folding problem, or given a protein structure, are scientists able
to identify the amino acid sequence that will fold into that structure?
2.2 Challenges in Modelling Novel Proteins
2.1 Modelling Novel Proteins As one can imagine, modelling proteins is a complicated task and scientists may
encounter many difficulties. One of the main problems is that protein mod-
A key idea when designing novel proteins is Anfisen’s dogma, which states that elling is time consuming. As stated in Levinthal’s Paradox, there are too many
the amino acid sequence determines the structure of the protein and that the possible conformations, which would take more time than the universe offers
protein’s shape is the thermodynamically favorable structure [n.andd]. As a to sample [VM18].There are 20 amino acids, resulting in 2n combinations for
matter of fact, there are numerous protein modelling softwares such as Rosetta a protein with n number of amino acids, and with the current time steps (in-
[n.andc] that are designed to generate a thermodynamically favorable structure crements of time in the simulation), it is not feasible to run through all the
for a protein. In general, proteins can be modelled based on the similarity of combinations. Continuing on with challenges in modelling proteins, for tem-
their tertiary structure, sequence (primary structure), or function with another plate based modelling, sometimes there may not be an available template to use
protein. As such, the two ways to model proteins are template based modelling, as a scaffold for modelling proteins and the accuracy of the modelling depends
in which an existing protein s tructure i s used as a s caffold f or modeling proteins, heavily on t he s imilarities between t he s equences. I n addition, t he f orce fields
153 154
used in de novo modelling may not be accurate and the shape outputted by the functions by inducing the binding of two IL-2 cell-membrane receptors, IL-2β
simulation may not resemble the actual protein shape in nature. Another issue (IL-2Rβ) and the common-gamma (γ or IL-2Rγ) receptor, which form the IL-
is that the number of conformations the protein can sample may be limited by 2Rβγ complex [TJR18]. Afterwards, a cascade of cell signaling is triggered
the energy levels in the simulation. Of course, there are other difficulties such in immune cells. However, some off-target cells have IL-2 alpha-subunits (IL-
as rotamer optimization [JMS00] and allosteric regulation that are not covered 2Rα or CD25) that can bind with the other two membrane receptors to form
in this text [AH15]. IL-2Rαβγ trimers, which have a greater affinity for binding to IL-2. There-
fore, the off-target cells are affected more than the target cells. Fortunately,
One way to obtain more accurate models of proteins is to create more efficient Neo-2/15 will not have this problem because the binding site and dependency
algorithms or increase computational power. In particular, volunteers can be of CD25 is eliminated, while the binding of IL-2Rβγ complex is still maintained.
used to speed up the calculations by spreading the computational burden over
many users. For example, volunteers can download the software Folding@Home Even still, creating novel therapeutic proteins has its difficulties, because as
[n.anda] and run simulations for protein folding on their own computer in the with any other drugs, it must be ensured that the protein does not provoke a
background. To address the inadequacies of the force fields, simulated anneal- severe immune response in a patient. In addition, the therapeutic proteins may
ing, which is raising the temperature of the simulation, can also be used to not be stable if a natural protein is used as a starting conformation. This is be-
allow the protein to fold into more possible conformations [AH15].Furthermore, cause natural proteins themselves are not stable, and changes such as amino acid
to address the challenge that the templates used in template modelling may substitutions when designing the protein may cause the protein to coagulate. So
only be distantly related, multiple templates may be used so that there are far, Neo-2/15 has been found to elicit a low immune response, and with de novo
more overlapping sequences, which will provide more accurate results [Zha08]. modelling, scientists can alter the hydrophobic interactions and other areas to
make the protein more stable. More proteins may be made based on the same
methods used to create Neo-2/15 to treat many diseases such as autoimmune
3 Applications of Novel Proteins diseases. In autoimmune diseases, the body attacks its own cells as it views
those cells forgein. But since Neo-2/15 resembles IL-2, which triggers regula-
There are countless areas that these novel proteins can be used after the protein tory T-cells that can suppress the immune system, proteins similar to Neo-2/15
structure is modelled and the protein is made, as proteins perform a variety may play a role in the development of a cure for autoimmune diseases [AQRS20].
of functions, such as digestion, transport, catalysis, contraction for muscles,
storage, protection, and structural support [KAT19]. Proteins are especially
In addition to their use in therapy, novel proteins can also be used to improve
relevant in biotechnology, which seeks to use biology to solve problems in the
or create new vaccines. Vaccines such as the yearly influenza (flu) vaccines are
world. Some of the major fields in biotechnology that novel proteins are used
crucial, and with novel proteins, David Baker’s group and researchers at the
in include the medical sector and the industrial sector.
National Institute of Allergy and Infectious Diseases’ Vaccine Research Center
(NIAID) are among those at the brink of creating a universal flu vaccine [Boy18]
3.1 Novel Proteins in the Medical Sector that may be effective for several years. A universal flu vaccine is important be-
cause flu strains rapidly mutate and it is difficult to predict which one will
Due to novel proteins, many new possibilities have been opened in fields such as
be more potent one year. The influenza strains are able to escape detection
immunology. Neoleukin-2/15 (Neo-2/15), developed by scientists at the Univer-
in the body because they change the shape of proteins on their surfaces called
sity of Washington, is one example of a de novo therapeutic protein, aimed at
antigens. Usually, the body neutralizes antigens from foreign substances by pro-
targeting cancer without producing side effects such as toxicity. The protein was
ducing proteins called antibodies binding to them. Vaccines exploit this idea
created using Rosetta and it resembles the cytokine interleukin-2 (IL-2), which
and display the antigens from a specific virus to stimulate a patient’s body to
controls the differentiation and homeostasis of both pro- and anti-inflammatory
produce antibodies for the antigens. However, since influenza mutates rapidly,
T cells by binding to receptors such as CD25 [RC18]. IL-2 is used in cancer
even if antibodies are created, it will not protect the body from the new strain.
treatment in the drug Proleukin, but IL-2 has a short window of effectiveness
and it may cause severe damage such as capillary leak syndrome when high
The universal flu vaccine resolves this by displaying the antigens from various
doses are needed because it may also affect healthy cells.
flu strains. The major antigen of influenza is hemagglutinin, which possesses a
globular head domain that mediates receptor binding and a stalk domain at the
Neo-2/15, on the other hand, will not produce these side effects because it
membrane-proximal region [Boy18]. When creating the universal flu vaccine,
enables activation of on-target tumor-fighting cells without preferentially acti-
scientists aim to target the stalk domain of hemagglutinin, a more conserved
vating t he off-target cells r esponsible f or t oxicity and i mmunosuppression. I L-2
region. The globular head can be removed by chemical methods and then neu-
155 156
tralizing antibodies for the stalk domain can be created [Kra15]. A challenge Fluorescent novel proteins such as the ones designed by Baker and Stoddard’s
to this is that the antibodies for the stalk domain may not be as effective as teams may be able to address the challenges associated with natural fluorescent
antibodies for the head. There are few antibodies for the stalk part because the proteins to improve their usage.
stalk domain evolves slower than the head region. Mutations in the stalk do not
spread rapidly and are therefore not usually detected [EKK18]. More research 3.2 Novel Proteins in the Industrial Sector
is needed but so far antibodies that target the stalk domain are promising.
Besides the medical field, industry is another major sector that novel proteins
Of course, having preventive measures to reduce the possibility of illness is opti- can be used in. They can be used to make a variety of new materials, such
mal. Novel proteins such as fluorescent proteins can be applied in diagnostics as as synthetic spider silk can be made, for example. Indeed, that is the goal of
well. There are intrinsically fluorescent proteins, which become fluorescent after many companies such as the Kraig Biocraft Laboratories because spider silk is
folding without addition of a fluorophore, and extrinsically fluorescent proteins strong, elastic, thin, and biodegradable [RS08]. As a comparison, spider silk
that must bind an endogenous molecule or ligand to fluoresce. These fluores- has a material toughness of 120,000 - 160,000 joules per kilogram (J/kg) while
cent proteins can be used to visualize cells, tissues, and organs; to monitor Kevlar has a material toughness of 30 - 50,000 J/kg and steel has a material
division and migration of cells in development, transplantology, inflammation, toughness of 2,000 - 6,000 J/kg [n.andf]. Spider silk’s properties are due to
and cancerogenesis; and to decipher neural circuits. However, it is uncertain if the highly repetitive structure of the proteins that make up spider silk. For
fluorescent proteins result in cellular toxicity because of the aggregation of the example, spider silk proteins contain many glycine residues, especially in their
fluorescent proteins or the generation of free radicals due to the excitation of cores. Spider silk however, cannot be mass produced by farming spiders be-
the fluorophores. There are other disadvantages, such as fluorescent proteins cause spiders are cannibalistic and territorial and they would have to be housed
may not be bright enough [Jen12]. Also, when fluorescent proteins are tagged individually, which is inefficient. In addition, an individual spider does not pro-
to other proteins to trace those proteins, fluorescent proteins may hinder other duce a large amount of silk, so harvesting silk from spiders would be impractical.
protein functions as fluorescent proteins are of a relatively big size.
Therefore, in order to acquire a large amount of spider silk for industrial pur-
Novel fluorescent proteins may overcome many of these challenges and in partic- poses, it has to be produced artificially. To do that, the first step is to get the
ular, David Baker’s team and Barry Stoddard’s team have developed a de novo proteins for spider silk by mass producing the silk proteins with recombinant
fluorescent protein, similar to Green Fluorescent Protein (GFP), that can bind methods in organisms such as E. coli or silk worms. After the silk proteins are
other molecules. Fluorescent proteins have a chromophore, which is the part produced, scientists can create spider silk by emulating the process that spiders
that allows for fluorescence. Chromophores consist of about 220 to 240 amino use to make silk. Spider silk is made up of proteins called spidroins, which are
acid residues which fold into a β-barrel formed by 11β-sheets that accommodate stored in a liquid state and at high concentration in the ampulla of the spinning
an internal distorted helix [DMCL10], and the researchers were able to model gland in the spider’s abdomen [CRN18] As the proteins pass through the spin-
the β-barrels and allow them to bind to target ligands using Rosetta [JDB18]. ning gland, they go through an acid environment and removal of water that will
This is a significant breakthrough because until recently, no β-barrel proteins solidify it into silk, which can then be excreted. In a similar way, spider silk can
were successfully modelled because β-strands tend to coagulate if they are not be made in labs by using a spinning apparatus. E. coli are first used to generate
perfectly aligned [Str18]. Furthermore, it is difficult to customize the proteins artificial proteins and the proteins are made in a liquid state. Afterwards, it
to bind to a specific target of interest due to the many possible orientations of goes through processes such as electrospinning, which is beyond the scope of
positions of a molecule in a protein cavity where ligands bind [LIM18]. this text, but in short, liquids go through a phase transition from liquid to solid
and fibers are formed. The fiber can then be spun into thread by a spinning
The researchers sought to create properly folding β-barrels and identified that wheel.
large local deviations in ideal β-strand twist were necessary to maintain continu-
ous hydrogen bond interactions between strands in the β-barrels. With this new Although it is promising so far, the fibers produced do not have the same quality
understanding, they were able to refine their methods to model the backbone, of natural spider silk yet. One of the reasons is that recombinant methods to
side chains, and other parts. And as some fluorescent proteins required bind- produce the proteins provide challenges. The length and size of the spider pro-
ing to other molecules such as ligands to function, they used a new algorithm teins make them difficult for bacterial hosts to synthesize and secrete, and for
which rapidly modelled the different areas that ligands could bind to (for their researchers to isolate and purify in solution. In addition, there is an incomplete
experiment they designed it to bind to a compound called DFHBI). The novel understanding of the underlying gene and/or amino acid sequences of most of
protein that they were able to design was found to bind to DFHBI with great the spider genes, so the proteins produced may tend to coagulate and function
affinity. I t also emitted greater fluorescence and was s maller t han GFP [ JDB18]. differently [ SJBF20]. But even s till, t he fibers produced are great progress.
157 158
Spider silk is a remarkable material and as methods to produce it continue to novo enzymes made that can perform catalytic functions for biological processes
improve, spider silk can be used to reinforce military gear or weapons and other is Syn-F4, designed by Hecht and others in 2011 [DeN]. At first, they designed
environmental friendly materials. de novo proteins that performed catalytic functions, in particular, a four-helix
protein called Syn-IF. They then experimented with E. coli and looked for genes
A different use of novel proteins in industry is water purification. Many mem- that would be lethal to the bacteria if the genes were removed, in order to test
brane proteins such as aquaporins have qualities namely high osmotic water the de novo proteins.
permeability and rejection of ions that are desirable for purifying water [IKB18].
In fact, some water filtration companies such as Aquaporin are using artificial The E. coli were modified so that they did not contain an enzyme called FeS.
aquaporins. Aquaporins are a type of transport protein that can be found in the E. coli acquires iron, an essential mineral for life, from their environment us-
cell membranes and they help facilitate the diffusion of water molecules into the ing a molecule called enterobactin, and FeS is needed to extract the iron from
cell. In particular, aquaporins are channel proteins and they have pores where enterobactin in order for the bacteria to use it. The E. coli without FeS were
water molecules can enter the cell. rendered almost dead without an iron supply and E. coli colonies began to turn
red as the iron that the bacteria could not use built up. That provided a perfect
Specifically, aquaporins have NPA motifs and aromatic/arginine (ar/R) regions environment for the researchers to test the abilities of the de novo proteins they
in their pores that give them the selectivity for water. NPA motifs are areas had and they used directed evolution to make newer de novo proteins that could
with the amino acids asparagine, proline, and alanine. These amino acids are extract iron better than Syn-F1. Syn-F1 did not show much enzymatic activity,
known to play a role in blocking protons from entering by creating a positive but they eventually hit upon one particular protein derived from Syn-IF, called
electrostatic field and breaking hydrogen bonds from water molecules, although Syn-F4, when it was added to the modified E. coli colonies, it turned the bac-
there is still controversy as to the specifics. There is also an ar/R region in teria into a healthy color again. This meant that Syn-F4 was able to extract
the pores with the amino acid arginine, which is positively charged. The ar/R iron from enterobactin just as FeS did and functioned as an enzyme.
region is known to form hydrogen bonds with water molecules, forcing the water
molecules to pass through aquaporin’s pore in a single file. It excluses ions and Syn-F4 is significant because it is a non-biological catalyst that can perform
is also narrow enough to exclude other molecules based on size. catalytic functions. It was not thought possible that something non-natural
could perform natural processes. In addition, it is challenging to create de novo
Artificial aquaporins can be incorporated into biosynthetic membranes that are enzymes for many reasons, for example, it is difficult to identify the optimal lo-
used in water filtration. These membranes can then be used to purify water cations on a protein for the active site and model the active site with sufficient
using forward or reverse osmosis. Even so, aquaporins are not stable enough accuracy to enable the protein to function appropriately [NK09]. Some ways
for industrial purposes when they are incorporated into the biosynthetic mem- scientists have tackled these difficulties are to use both computational mod-
branes, as chemicals such as detergent may denature the proteins. One way to elling and experimental methods. Many attempts have been made to improve
increase the stability is to place vesicles that naturally have aquaporins into the the computational methods, which are too detailed to be covered in this paper,
synthetic membranes. More research is also being done to improve the aqua- and one such example is a new Rosetta platform that the Baker Lab has used
porins as they cannot filter small ions such as sodium ions that well yet and to design de novo enzymes [FRB11]. Once a model for the protein is created,
therefore cannot be used for desalination (filtering salt water), which is impor- proteins can be refined with methods such as directed evolution, similar to how
tant in places that have a lack of access to fresh water. Lack of safe water in Syn-F4 was made. Certainly, many chemical functions can be performed as
places such as rural Africa is still an issue and it would be a huge advantage if more artificial enzymes are created.
the production of de novo aquaporins could be refined.
Another significant way novel proteins can be useful in industry is the use of 4 Conclusion
novel enzymes. Many biological functions would be impossible without enzymes,
These novel proteins that scientists have created can be widely used for medical
which are proteins that speed up chemical reactions necessary for life. Many
purposes such as therapy, vaccine development, and diagnosis, or even for in-
chemical reactions such as those in cellular respiration would take far too long
dustrial purposes such as the creation of new biomaterials, water filtration, and
to happen on their own for us to survive. As a brief description of enzymes
chemical processes with enzymes. There are endless possibilities that these new
function, they bind to substrates (the substance the enzyme is acting on and
proteins can offer, and an essential part of designing these proteins is modelling
can be thought of as reactants in a chemical reaction) at active sites and convert
them, as mentioned earlier in the paper. But whether template modelling or de
them into products. Since enzymes are very important and it is not surprising
novo modelling i s used, generating t he s tructure of t he protein r equires s ignifi-
that s cientists have worked hard t o create s ynthetic enzymes. One of t he first de
159 160
cant computational power and data. Furthermore, the protein models generated [Bak17] David Baker. Unleashing the power of synthetic proteins. Science
are not always accurate. There are also added difficulties when synthesizing the Philanthropy Alliance, March 2017.
proteins, such as ensuring the proteins are stable and properly functioning.
[Bak19] David Baker. What has de novo protein design taught us about
In general, it is a long process for proteins to go from the laboratory to the protein folding and biophysics? Wiley Online Library, February 12,
market, one major example being that vaccine development often takes years 2019.
to complete. After sequencing the virus’s genetic material, a protein for the [BBB20] Asim K. Bera Christoffer Norn Cameron M. Chow Lauren P. Carter
vaccine has to be created and optimized, which involves much research. Once Inna Goreshnik Frank Dimaio Benjamin Basanta, Matthew J. Bick
the vaccine is created, it has to go through numerous clinical trials before it can and David Baker. An enumerative algorithm for de novo design of
be used. Improving the accuracy and speed of the protein models predicted is proteins with diverse pocket structures. PNAS, September 8, 2020.
one possible way to speed up the process of vaccine development. Going back
to the example of the vaccine, using computational methods to understand the [Boy18] Alan Boyle. Open philanthropy project awards 11m to protein de-
virus’s structure and how a new drug interacts with it can help scientists to signers for universal flu vaccine. Geek Wire, April 4, 2018.
refine the protein and get it to clinical trials sooner. It would be to our advan-
tage to find ways to develop these novel proteins faster so that they can be used. [CRN18] Jessica P. Bunz Charlotte Rat, Julia C. Heiby and Hannes
Neuweiler. Two-step self-assembly of a spider silk molecular clamp.
Indeed, this is perfectly summed up by Ryan Bethencourt’s words, “Our world is Nature, November 14, 2018.
built on biology and once we begin to understand it, it then becomes a technol- [CWP15] Christine Tinberg Alessandra Camarca Carmen Gianfrani Shirley
ogy.” Proteins can be used to create things that benefit us such as medicines. Paski Rongjin Guan Gaetano Montelione David Baker Clancey Wolf,
For example, the synthetic protein Kuma030 [CWP15] can be used in devel- Justin B. Siegel and Ingrid S. Pultz. Engineering of kuma030: A
oping a cure for celiac disease, a serious autoimmune disease in which people gliadin peptidase that rapidly degrades immunogenic gliadin pep-
cannot eat gluten because the ingestion of gluten can trigger an inflammatory tides in gastric conditions. American Chemistry Society, September
response in their intestines [n.a20]. And even besides the medical and industrial 15, 2015.
sector discussed in this paper, novel proteins have many uses such as in protein
switches [Bak17] and biosensors [CWP15] and even for waste management in [DeN] A de novo designed metalloenzyme for the hydration of co2 , au-
space. Many things seem to naturally work so well, so why not emulate or refine thor=Dr. Virginia M. Cangelosi, Dr. Aniruddha Deb, Prof. James E.
them? With the current progress scientists are making, imagine what will be Penner-Hahn, and Prof. Vincent L. Pecoraro, year=June 18, 2014,
possible in the future. journal=Wiley Online Library,.
[DMCL10] Sergey Lukyanov Dmitriy M. Chudakov, Mikhail V. Matz and Kon-
References stantin A. Lukyanov. Fluorescent proteins and their applications in
imaging living cells and tissues. American Physiology Society, July
[AH15] Modesto Orozco Josep L Gelpı́ Adam Hospital, Josep Ramon Goñi. 1, 2010.
Molecular dynamics simulations: advances and applications. Dove
Press, July 27, 2015. [EKK18] Patrick C. Wilson Justin Bahl Ericka Kirkpatrick, Xueting Qiu and
Florian Krammer. The influenza virus hemagglutinin head evolves
[AQRB20] Jooyoung Park Hansol Lee Robert A. Langan Scott E. Boyken Marc faster than the stalk domain. Nature, July 11, 2018.
J. Lajoie Longxing Cao Cameron M. Chow Marcos C. Miranda Jimin
Wi Hyo Jeong Hong Lance Stewart Byung-Ha Oh Alfredo Quijano- [Fis10] Andras Fiser. Template-based protein structure modeling. Springer
Rubio, Hsien-Wei Yeh and David Baker. De novo design of modular Link, August 17, 2010.
and tunable allosteric biosensors. US National Library of Medicine
[FRB11] Sagar D. Khare Sinisa Bjelic Florian Richter, Andrew Leaver-Fay
National Institutes of Health, July 20, 2020.
and David Baker. De novo enzyme design using rosetta3. PLOS,
[AQRS20] Carl D.Walkey Alfredo Quijano-Rubio, Umut Y.Ulge and Daniel- May 16, 2011.
Adriano Silva. The advent of de novo proteins for cancer im-
[Hev09] Dennis Hevesi. Ralph f. hirschmann, leading scientist on early en-
munotherapy. Science Direct, June 2020.
zyme research, dies at 87. The New York Times, July 18, 2009.
161 162
[IKB18] Yves Marie Legrand Istvan Kocsis, Zhanhu Sun and Mihail Bar- [n.andf] n.a. Spidersilk. Kraig Biocraft Laboratories, Inc., n.d.
boiu. Artificial water channels—deconvolution of natural aquaporins
through synthetic design. Nature, August 1, 2018. [NK09] Vikas Nanda and Ronald L. Koder. Designing artificial enzymes by
intuition and computation. Nature, December 17, 2009.
[JDB18] William Sheffler Lindsey A. Doyle Hahnbeom Park Matthew J. Bick
Binchen Mao Glenna W. Foight Min Yen Lee Lauren A. Gagnon [PDRK12] Richard Bonneau P. Douglas Renfrew, Eun Jung Choi and Brian
Lauren Carter Banumathi Sankaran Sergey Ovchinnikov Enrique Kuhlman. Incorporation of noncanonical amino acids into rosetta
Marcos Po-Ssu Huang Joshua C. Vaughan Barry L. Stoddard Ji- and use in computational protein-peptide interface design. PlOS
ayi Dou, Anastassia A. Vorobieva and David Baker. De novo design One, March 14, 2012.
of a fluorescence-activating β-barrel. Nature, September 12, 2018. [PX09] Jian Peng and Jinbo Xu. Boosting protein threading accuracy.
[Jen12] Ellen C. Jensen. Use of fluorescent probes: Their effect on cell biol- Springer Link, 2009.
ogy and limitations. American Association For Anatomy, October [RC18] Sarah H. Ross and Doreen A. Cantrell. Signaling and function of
12, 2012. interleukin-2 in t lymphocytes. Annual Reviews, April 2018.
[JLZ17] Peter L. Freddolino Jooyoung Lee and Yang Zhang. Ab initio protein [RECZ12] Ran Chao Ryan E. Cobb and Huimin Zhao. Directed evolution:
structure prediction. Springer Link, April 13, 2017. Past, present, and future. AIChE Journal, December 12, 2012.
[JMS00] Maria Arménia Carrondo Joaquim Mendes, António M. Baptista [RS08] Lin Römer and Thomas Scheibel. The elaborate structure of spider
and Cláudio M. Soares. Improved modeling of side-chains in pro- silk. US National Library of Medicine National Institutes of Health,
teins with rotamer-based methods: A flexible rotamer model. Wiley October-December 2008.
Online Library, January 25, 2000.
[SJBF20] Patrick T. Spicer Sean J. Blamires and Patricia J. Flanagan. Spi-
[KAT19] Indira Rajagopal Kevin Ahern and Taralyn Tan. Structure function- der silk biomimetics programs to inform the development of new
proteins i. Biolibre Texts, June 23, 2019. wearable technologies. Frontiers In Materials, February 18, 2020.
[KB19] Brian Kuhlman and Philip Bradley. Advances in protein structure [Str18] Rita Strack. Fluorescent proteins from scratch. Nature, October 30,
prediction and design. Nature Journal, August 15,2019. 2018.
[Kra15] Florian Krammer. The quest for a universal flu vaccine: Headless [TJR18] Caicun Zhou Tao Jiang and Shengxiang Ren. Role of il-2 in cancer
ha 2.0. Cell Host and Microbe, October 14, 2015. immunotherapy. Taylor and Franis Online, April 2018.
[LIM18] Y LIM. Barreling the way to designer proteins. Fred Hutch, Novem- [VM18] Alexey V.Melkikh and Dirk K.F. Meijer. On a generalized levinthal’s
ber 19, 2018. paradox: The role of long- and short range interactions in complex
[MM05] Nina M.Antikainen and Stephen F. Martin. Altering protein speci- bio-molecular reactions, including protein and dna folding. Science
ficity: techniques and applications. Wiley Online Library, April 15, Direct, January 2018.
2005. [Zha08] Yang Zhang. Progress and challenges in protein structure prediction.
[n.a20] n.a. Celiac disease. Medline Plus, August 18, 2020. Science Direct, June 2008.
[n.anda] n.a. About folding@home. Folding@home, n.d.
[n.andb] n.a. Designing proteins from scratch. Amgen Science, n.d.

[n.andc] n.a. Getting started. Rosetta Commons, n.d.
[n.andd] n.a. Protein folding and the thermodynamic hypothesis, 1950-1962.
U.S. National Library of Medicine, n.d.
[n.ande] n.a. Rcsb pdb: Homepage. RCSB PDB, n.d.
163 164
stimuli with memories, thereby demonstrating its important function in learning
and memory [Gil15].
In order to understand cognition, neurogenesis, and the molecular mecha-
nisms that mediate the neural underpinnings of aerobic exercise’s benefits, this
A Deconstructive Approach to Hippocampal review will take a deconstructive approach, starting with cognition and hip-
Neurogenesis as a Function of Aerobic Exercise pocampal volume. It will then explore a mediating neurotrophic factor, BDNF,
that may be involved in this process and influence hippocampal neurogenesis.
∗
At the end, future directions to better understand the effects of aerobic exercise
Harshitha Valluri on these processes are discussed.
April 2, 2021
2 Hippocampal Neurogenesis in Humans
There has been controversy over whether hippocampal neurogenesis occurs in
Abstract adult humans [Mau18]; [Sha18]. One paper suggests that neurogenesis continues
Aerobic exercise, specifically running, leads to neural benefits in ro- through aging by looking at cells that are positive for the markers DCX/PSA-
dents and humans. In this review, a deconstructive approach is taken to NCAM, which label immature granule neurons. DCX/PSA-NCAM+ neurons
understand the neural benefits of aerobic exercise in terms of cognition, were found to be consistent in number throughout age and were seen in at least
hippocampal size, brain-derived neurotrophic factor (BDNF), and hip- the thousands in the anterior, mid, and posterior dentate gyrus [Mau18]. An-
pocampal neurogenesis. These four benefits were positively impacted as a other paper, which looked at the same cell marker in immature neurons in the
result of running in rodents and the same was seen in humans except for various layers of the dentate gyrus, the molecular zone, subgranular zone, gran-
with hippocampal neurogenesis. Human hippocampal neurogenesis has ular cell layer, and hilus, determined that hippocampal neurogenesis decreases
been demonstrated to continue throughout the lifespan but there was not in children to minuscule levels in adults [Sha18]. Both papers also looked at
evidence for it as a result of running. Therefore, future directions address other markers such as the protein Ki67 for proliferating cells, the transcription
potential ways to study hippocampal neurogenesis in humans in relation
factor SOX2 and protein Nestin for early neural progenitors, and the protein
to aerobic exercise. The future directions also discuss potential dosage
curves in regards to intensity, duration, and frequency of aerobic exercise
NeuN for mature neurons [Mau18]; [Sha18]. In [Mau18], Sox2+ cells declined
on cognition, hippocampal size, and BDNF. throughout aging but Ki67+ cells, Nestin+ cells, and NeuN+ cells remained
stable. In [Sha18], Ki-67+Sox2+ cells decreased in the first years in almost all
cases.
1 Introduction There could be multiple reasons that could explain the discrepancies be-
tween the two studies, but one outstanding factor is related to the brains that
For a long time, it has been known that exercise and cognition are positively were used. [Mau18] used intact brains whereas [Sha18] used brains of varying
connected [Rob94]; [Art99]. Recently, there has been a rise in people taking up neurological conditions including twelve subjects with epilepsy. The reasons
running as an independent form of aerobic exercise [Yan19]; [Jac19]. Scientists for death were various and not explicitly given for the subjects who died at 14
have recently started to elucidate the neural underpinnings that may mediate gestational weeks up to death at 77 years old in [Sha18]. Another factor could
this positive relationship between aerobic exercise and cognition. While the be the postmortem interval which was between 4 to 26 hours in [Mau18] but
neural mechanisms are more clear in rodents than they are in humans, studies in [Sha18], it was 48 hours or less.
have shown that hippocampal neurogenesis, the birth and development of new Two more recent papers both show that adult hippocampal neurogenesis
neural cells [Pet98], is a likely mediator of the relationship between aerobic continues in older human brains [Mat19]; [Ele19]. Unlike the other two pa-
exercise and enhanced cognition [Hen99a]; [Hen05]. pers, these studies did not look at hippocampal neurogenesis in people of all
In humans, the hippocampus is one brain region in which there has been ages but focused on a specific age group of 79 to 99 in [Mat19], 43 to 87 for the
evidence that neurogenesis occurs [Ola15]. The hippocampus is one of the major healthy brains, and 52 to 97 for subjects with Alzheimer’s disease [Ele19]. While
parts of the brain which is involved with the cognitive functions of learning both studies found persistent hippocampal neurogenesis in healthy older adults,
and memory [Kul12]. One part of the hippocampus that has been the focus of they did not find the same results for people with Alzheimer’s disease. [Ele19]
hippocampal neurogenesis research, is the dentate gyrus, which connects sensory looked for DCX/PSA-NCAM+ cells and found that immature neurons declined
∗ Advised by: Patrick Liu of the University of Oxford
as Alzheimer’s disease progressed in the brain and were in lower counts com-
165 166
pared to healthy brains. [Mat19] looked at DCX+PCNA+ immature neurons 3.2 Hippocampal Volume and Density Likely Increases as
and found that hippocampal neurogenesis was present in people with mild cog- a Result of Aerobic Exercise
nitive impairment and Alzheimer’s disease. While there has been controversy
about neurogenesis continuing past the first years of life in humans, overall, In a study on mice with Alzheimer’s disease, it was found that mice who ran
more recent studies suggest that neurogenesis does continue in healthy peo- had increased hippocampal volumes, as measured by Nissl staining, compared
ple [Mau18]; [Ele19]; [Sha18]; [Mat19]. Neurogenesis has been further studied to mice that did not run. This was the case in both mice that voluntarily
in rodents, and evidence suggests that neurogenesis continues throughout life exercised on a wheel as well as mice that were forced to run on a treadmill
and aerobic exercise enhances neurogenesis [Yi-18]; [Tae18]; [Hen99a]; [Hen99b]; [Car09]. Multiple studies in rats have also found increased hippocampal density
[Hen05]. of neurons resulting from aerobic exercise [Edu19]; [Xia16].
3.3 Aerobic Exercise Increases BDNF

3 Aerobic Exercise’s Neural Impact in Rodents
BDNF aids with neuronal development, particularly with differentiation, mat-
3.1 Aerobic Exercise Improves Cognition uration, and survival [Sir15], and it has been found to be required for neu-
rogenesis [Chi06]. Aerobic exercise increased BDNF, measured using Western
Aerobic exercise does enhance cognition in young adult, middle aged, and aged blotting, in the hippocampus [Tae18]. A few days of voluntary wheel running
mice [Yi-18]; [Tae18]; [Hen99a]; [Hen05]. A study with young rats found that has been found to result in increased BDNF mRNA levels in the hippocam-
aerobic exercise improved spatial learning as seen by decreased path length and pus [S. 95]; [Car02]. The increase of BDNF mRNA after aerobic exercise has
latency when finding the platform in the Morris water maze task, a widely been seen to last throughout just one week of voluntary exercise [S. 96]; [Sho03])
used spatial learning task for rodents [Nas03]. It has also been observed that as well as over the course of several weeks [Ame99]. Hippocampal BDNF protein
young adult mice who run voluntarily, compared to those that did not run, levels were also increased in young mice directly after voluntary wheel running
have improved spatial learning [Hen99a]. This increase in spatial learning has for three weeks and, following the three weeks of running, BDNF protein levels
also been seen using the Radial arm water maze, a hybrid Morris water maze began decreasing after one and two weeks and returned to baseline levels by
type task which can also determine errors [Nic10]. This study also found that weeks three and four [Nic10]. While BDNF increase as a result of aerobic exer-
the young mice who ran voluntarily and were trained on the Radial arm water cise has mainly been seen in males, it is also true that females who voluntarily
maze 1 week after aerobic exercise did better on the maze than right after aerobic run have increased levels of BDNF mRNA and BDNF protein [Nic01]. BDNF
exercise or 2 weeks after aerobic exercise. In middle aged mice, it was found has been shown to mediate the positive effects of aerobic exercise on cognition
that spatial learning is better improved in mice that did intermittent exercise, directly [Sho04]. When BDNF action was blocked in exercising rats, their spa-
which means voluntary running for 2 weeks and no running for 1 week being tial learning, as measured by the Morris Water Maze Task, dropped to the levels
repeated for a total of 12 weeks, compared to continuous voluntary exercise and of the control group.
no exercise [Yi-18]. Similarly, this finding was consistent in a study with 1 hour While BDNF expression has been linked to improvements in cognitive func-
of daily forced treadmill aerobic exercise with mild exercise (15 m/min) being tioning, there is a mechanism that demonstrates how exercise can increase
better at improving spatial learning than intense exercise (40 m/min) or no BDNF expression. Exercise, specifically aerobic exercise, stimulates the in-
exercise [Kos15]. In aged mice, voluntary wheel running increased their spatial creased expression of the hippocampal gene Fndc5 (a glycosylated type I mem-
learning as well [Hen05]. In older mice, the magnitude of change in latency brane protein) in the brain. The gene is stimulated by the transcriptional reg-
and path length is greater between older runners and older control mice than ulators Pgc1a and Erra which are induced by skeletal muscles as a response to
between young runners and young controls. The older runner mice’s latency aerobic exercise. As a result of the increased Fndc5 gene expression, BDNF
and path length decreased to similar amounts as in young mice suggesting that expression also increases which consequently will lead to better cognitive func-
older age may not restrict aerobic exercise’s improvements on spatial learning. tioning due to the positive effects of BDNF in the hippocampus.
Forced treadmill running also had positive cognitive improvements on short
term memory, which was assessed using the step-down avoidance task, as well
as spatial learning [Tae18]. Another study looking at rats found that specifically 3.4 Aerobic Exercise Enhances Hippocampal Neurogene-
lower intensity forced aerobic exercise (15 m/min for 40 minutes a day) had more sis
beneficial effects on short term delayed working memory, which was assessed Running enhances hippocampal neurogenesis in young adult, middle aged, and
using the delayed spatial alternation task, compared to mid (20 m/min) and aged mice. [Yi-18]; [Tae18]; [Hen99a]; [Hen99b]; [Hen05]. Hippocampal neuro-
high intensity aerobic exercise (30 m/min) [Xia16]. genesis in mice has been measured using cells that were at least double marked
167 168
for BrdU, which is a marker for proliferating cells, and NeuN [Tae18]; [Hen99a]; levels immediately after aerobic exercise in young males who did acute exer-
[Hen99b]; [Hen05]. It has also been measured using anti-BrdU, which are the cise and chronic exercise for both 3 and 5 weeks, 60 minutes to 90 minutes of
antibodies marking BrdU positive cells [Yi-18]. This study also found that acute exercise resulted in an increase in serum BDNF for the groups [Ead11].
intermittent aerobic exercise in middle-aged mice was better at increasing hip- Similarly, a study in young, healthy men also found increased serum BDNF
pocampal neurogenesis compared to continuous aerobic exercise and no aerobic after moderate intensity aerobic exercise compared to before training [Jer08]. A
exercise. Further, one study found that mild long term forced aerobic exer- study looking at healthy, older people between the ages of 65 and 80 found that
cise was better at enhancing hippocampal neurogenesis than intense long term serum BDNF was elevated after a 35 minute session of aerobic exercise [Kri17].
aerobic exercise over a period of six weeks [Kos15]. BDNF amounts have been seen to increase by over 3 fold after aerobic exer-
cise [Cam14]. One study found that from 20 minutes of vigorous aerobic exercise
to 40 minutes there was a 2.7 times increase and a 1.4 times increase for the
4 Aerobic Exercise’s Neural Impact in Humans moderate aerobic exercise group [Mat13]. This suggests that aerobic exercise
intensity affects the magnitude of BDNF increase.
4.1 Aerobic Exercise Enhances Cognition
Aerobic exercise has been shown to improve cognitive function in humans [Art99];
[Cha08]. In children, aerobic fitness positively impacts cognitive control, as-
5 Conclusions and Future Directions
sessed by the flanker task (a selective attention paradigm), likely through in- Evidence for aerobic exercise related neurogenesis in humans is currently lack-
creases of brain volume in areas like the hippocampus [Lau10b]. A clinical trial ing, but it has been clearly demonstrated in rodents. It has been observed in
in young adults found that executive functioning, which was measured using set both rodents and humans that aerobic exercise enhances cognition, hippocam-
switching between tasks and The Groton Maze Learning Test, improved with pal size, and BDNF. In general, neurogenesis in humans has still been disputed
aerobic exercise as well [Yaa19]. Another clinical trial in older adults, which over the past few years with more recent evidence suggesting it continues into
consisted of a cross-sectional study and longitudinal study of 6 months, found adulthood. This section will first discuss aerobic exercise’s benefits on cogni-
that higher cardiovascular fitness training is connected to improved executive tion, hippocampal size, and BDNF levels in regards to a dosage curve and at
functioning performance as assessed by the flanker task [Sta04]. A more recent the end of this section, future directions in this field to help better understand
meta-analysis of 39 studies also found that moderate or higher intensity aerobic the relationship between aerobic exercise and neurogenesis are discussed.
exercise of 45 to 60 minutes improved cognition of adults over 50 years old no One main area in humans that is lacking knowledge is regarding a potential
matter their initial cognitive status [Jos18]. dosage curve between aerobic exercise and cognition, hippocampal size, and
BDNF increase, as well as potential age-dependent effects. Determining the
4.2 Aerobic Exercise Increases Hippocampal Volume dosage curve shape (inverted U-shaped, J-shaped, etc.) would help show what
duration, intensity, and frequency of aerobic exercise is needed to get the best
Aerobic fitness and spatial learning are both positively associated with hip-
benefits for cognition, hippocampal volume, and BDNF increase while looking at
pocampal volume [Kir09]. Aerobic exercise specifically has been shown to in-
age could help people personalize their exercise regimes for optimal neurological
crease hippocampal volume in both the left and right hippocampus as seen by
benefit. Conducting a study regarding dosage curves in rodents as well could
MRI scans in elderly human adults [Kir11]. On the opposite age spectrum,
help show connections between cognition, hippocampal volume, BDNF, and
comparing preadolescent children, those who have higher aerobic exercise lev-
neurogenesis so that the nature of their relationship could be extended to better
els have increased hippocampal volumes as well [Lau10a]. For early and middle
understand neurogenesis in humans.
aged adults, the findings were similar in a study where physical activity was self-
A study in rodents that could show the connection between aerobic exercise
reported and results showed that higher hippocampal volume positively corre-
benefits and dosage curve in humans would consist of healthy rodents of four
lated to aerobic exercise amount [Wil13]. The mechanism underlying increased
various age groups: adolescents, young adults, middle-aged, and elderly. For
hippocampal volume is the positive change in tissue density, which in turn in-
healthy humans, there would be five various age groups: preadolescents, adoles-
creases hippocampal volume [Mai16].
cents, young adults, middle-aged adults, and elderly adults. Preadolescents are
excluded from being an age group tested in rodents since mice and rats do not
4.3 Aerobic Exercise Results in BDNF Increase open their eyes until adolescence which begins at 2 weeks old [Abi17]; [Ste02].
The age groups in rodents and humans would be split into seven groups that
A study on healthy college men found increased plasma, platelet, and serum
would have to do aerobic exercise for either 0, 30, 60, 90, 120, 150, or 180 min-
BDNF after exercising on a treadmill in comparison to their levels at rest be-
utes. These groups would also be s plit t o account f or t hree different i ntensities
fore the aerobic exercise [Hyu12]. In another study that also looked at BDNF
169 170
of aerobic exercise (low, moderate, and high) as determined by percent of heart pocampus or if volume increases as a result of periphery neurons and expansions
rate reserve. They would be further split, to account for frequency, into groups occurring. It could also potentially rule out the effects of angiogenesis.
that exercise for 1, 2, 3, 4, 5, 6, or 7 days a week. Since BDNF directly supports neurogenesis [Chi06] and BDNF injections’
Before and after each group does their treadmill exercise, rodents would neuronal benefits have an inverted U-shape [Lau00], it may be likely that neuro-
complete the Morris water maze task to assess spatial learning and the step- genesis in rodents would follow this inverted U-shape as well. In rodents, mod-
down avoidance task to assess short term memory. To look at hippocampal erate intensity and 3 to 5 days would likely be best for neurogenesis since mild
size, specifically volume and density, they would also get MRI scans and Nissl aerobic exercise [Kos15] and intermittent aerobic exercise [Yi-18] were found to
staining, which stains neurons [And09]. Humans would complete the flanker task be most beneficial. The results could be extended to better understand the rela-
to look at cognition and get MRI scans to look at hippocampal size before and tionship between aerobic exercise and neurogenesis in humans. To fully test the
after exercise. Both rodents and humans would also have their plasma, platelet, possibility of exercise induced hippocampal neurogenesis in humans, the best
and serum BDNF measured to determine a potential dosage curve between way to do so would be to find out a noninvasive way of assessing the levels of
aerobic exercise and BDNF. In rodents, one further aerobic exercise benefit immature neurons and other neurons in the developmental stage. The reason
that would be tested is neurogenesis. Before aerobic exercise, BrdU would be why a noninvasive way is needed is because one of the only other ways to test
injected to measure proliferating brain cells, after aerobic exercise DCX/PSA- this would be by testing the effects of aerobic exercise in terminal patients or
NCAM would be used to mark immature neurons and NeuN would be used to via unethical routes; therefore, the main way to do this would be to develop
mark mature neurons. Measuring the amount of these types of neurons after new techniques to measure neurogenesis in vivo and noninvasively.
aerobic exercise every day until the fluctuation of neuron amounts levels out and
then comparing the amounts to the individuals’ baselines could help determine
a dosage curve relationship between aerobic exercise and neurogenesis. 6 Acknowledgements
The results of this study could likely show an inverted U-shaped curve or
exponential curve for cognition, hippocampal size, and BDNF increase in all ages I would like to sincerely thank Mr. Patrick Liu for all his guidance and help-
of rodents and humans since injecting BDNF has been observed to show inverted ful feedback throughout the research process. I would also like to thank Mr.
U-shape for neuronal regeneration in rodents [Lau00] and since around an hour David Ju for his useful editing assistance as well as Horizon Academic for this
or more of aerobic exercise has been shown to increase BDNF in humans [Ead11]. opportunity.
For age dependence, this study could likely show that older adults and rodents
can still gain the benefits of improved cognition, increased hippocampal size, References
and heightened BDNF levels that younger adults and rodents do since this was
seen with cognition before in mice [Hen05]. Moderate intensity will likely be [Abi17] Abigail E. Agoglia, and Sarah E. Holstein, and Amanda T. Small, and
best for these three benefits as observed in young men for BDNF levels [Jer08]. Marina Spanos, and Brainard M. Burrus, and Clyde W. Hodge. Com-
Around 3 to 6 days a week of aerobic exercise would likely be best for cognition, parison of the adolescent and adult mouse prefrontal cortex proteome.
hippocampal size, and heightened BDNF levels since 3 days a week has been PLOS ONE, 2017.
observed to improve cognition in humans [Sta04]; [Yaa19] and 6 days a week
did the same in mice [Tae18]. Overall, these results would likely demonstrate a [Ame99] Amelia Russo-Neustadt, and Ryan C. Beard, and Carl W. Cotman.
duration of 1 to 2 hours of moderate intensity aerobic exercise done 3 to 6 days a Exercise, antidepressant medications, and enhanced brain derived
week to be most beneficial to humans regardless of age. Having this information neurotrophic factor expression. Neuropsychopharmacology, 1999.
could help humans know how to best support their brain health through their
[And09] Andrea Kádár, and Gábor Wittmann, and Zsolt Liposits, and Csaba
exercise regime.
Fekete. Improved method for combination of immunocytochemistry
The results regarding hippocampal size, specifically volume, in rodents and
and nissl staining. Journal of Neuroscience Methods, 2009.
humans will likely be similar as in the mice with Alzheimer’s [Car09] and the
mechanism behind it in rodents will likely correspond with humans [Mai16] [Art99] Arthur F. Kramer, and Sowon Hahn, and Neal J. Cohen, and Marie T.
while potentially showing redundancy in the mechanism. The MRI scans and Banich, and Edward McAuley, and Catherine R. Harrison, and Julie
Nissl staining of the hippocampus could help find out whether aerobic exercise Chason, and Eli Vakil, and Lynn Bardell, and Richard A. Boileau,
increases hippocampal density, hippocampal volume or both, as well as if the and Angela Colcombe. Ageing, fitness and neurocognitive function.
specific mechanism through which this occurs in rodents is the same as in hu- Nature, 1999.
mans. This study could provide important results regarding whether density
increases as a result of connections forming within existing circuits in the hip-
171 172
[Cam14] Cameron S. Mang, and Nicholas J. Snow, and Kristin L. Campbell, hippocampal neurogenesis is abundant in neurologically healthy sub-
and Colin J. D. Ross, and Lara A. Boyd. A single bout of high- jects and drops sharply in patients with alzheimer’s disease. Nature
intensity aerobic exercise facilitates response to paired associative Medicine, 2019.
stimulation and promotes sequence-specific implicit motor learning.
Journal of Applied Physiology, 2014. [Gil15] Gilian F. Hamilton, and Justin S. Rhodes. Chapter sixteen - exercise
regulation of cognitive function and neuroplasticity in the healthy
[Car02] Carl W. Cotman, and Nicole C. Berchtold. Exercise: a behavioral and diseased brain. Progress in Molecular Biology and Translational
intervention to enhance brain health and plasticity. Trends in Neuro- Science, 2015.
sciences, 2002.
[Hen99a] Henriette van Praag, and Brian R. Christie, and Terrence J. Se-
[Car09] Carla M Yuede, and Scott D Zimmerman, and Hongxin Dong, and jnowski, and Fred H. Gage. Running enhances neurogenesis, learn-
Matthew J Kling, and Adam W Bero, and David M Holtzman, and ing, and long-term potentiation in mice. Proceedings of the National
Benjamin F Timson, and John G Csernansky. Effects of voluntary Academy of Sciences of the United States of America, 1999.
and forced exercise on plaque deposition, hippocampal volume, and
behavior in the tg2576 mouse model of alzheimer’s disease. Neurobi- [Hen99b] Henriette van Praag, and Gerd Kempermann, and Fred H. Gage. Run-
ology of Disease, 2009. ning increases cell proliferation and neurogenesis in the adult mouse
dentate gyrus. Nature Neuroscience, 1999.
[Cha08] Charles H. Hillman, and Kirk I. Erickson, and Arthur F. Kramer. Be
smart, exercise your heart: exercise effects on brain and cognition. [Hen05] Henriette van Praag, and Tiffany Shubert, and Chunmei Zhao, and
Nature Reviews Neuroscience, 2008. Fred H. Gage. Exercise enhances learning and hippocampal neuroge-
nesis in aged mice. The Journal of Neuroscience, 2005.
[Chi06] Chiara Rossi, and Andrea Angelucci, and Laura Costantin, and
Chiara Braschi, and Mario Mazzantini, and Francesco Babbini, and [Hyu12] Hyun-chul Cho, and Jongkyu Kim, and Sungyeon Kim, and Yeon
Maria Elena Fabbri, and Lino Tessarollo, and Lamberto Maffei, and Hee Son, and Namju Lee, and Seung Ho Junge. The concentrations of
Nicoletta Berardi, and Matteo Caleo. Brain-derived neurotrophic fac- serum, plasma and platelet bdnf are all increased by treadmill vo2max
tor (bdnf) is required for the enhancement of hippocampal neuroge- performance in healthy college men. Neuroscience Letters, 2012.
nesis following environmental enrichment. European Journal of Neu-
[Jac19] Jack Spittler, and Lauren Oberle. Current trends in ultramarathon
roscience, 2006.
running. Current Sports Medicine Reports, 2019.
[Chr13] Christiane D. Wrann, and James P. White, and John Salogiannnis,
[Jer08] Jerzy A. Zoladz, and Andrzej Pilc, and Joanna Majerczak, and Marcin
and Dina Laznik-Bogoslavski, and Jun Wu, and Di Ma, and Jiandie D.
Grandys, and Justyna Zapart-Bukowska, and K. Duda. Endurance
Lin, and Michael E. Greenberg, and Bruce M. Spiegelman. Exercise
training increases plasma brain-derived neurotrophic factor concentra-
induces hippocampal bdnf through a pgc-1α/fndc5 pathway. Cell
tion in young healthy men. Journal of Physiology and Pharmacology:
Metabolism, 2013.
An Official Journal of the Polish Physiological Society, 2008.
[Ead11] Eadaoin W. Griffin, and Sinéad Mullally, and Carole Foley, and Stuart
[Jer15] Jeremy Young, and Maaike Angevaren, and Jennifer Rusted, and Naji
A. Warmington, and Shane M. O’Mara, and Áine M.Kellyac. Aero-
Tabet. Aerobic exercise to improve cognitive function in older people
bic exercise improves hippocampal function and increases bdnf in the
serum of young adult males. Physiology & Behavior, 2011. without known cognitive impairment. Cochrane Database of System-
atic Reviews, 2015.
[Edu19] Eduardo Varejão Dı́az Placencia, and Fernando Tadeu Serra, and Jes-
sica Salles Henrique, and Ricardo Mario Arida, and Sérgio Gomes da [Jos18] Joseph Michael Northey, and Nicolas Cherbuin, and Kate Louise
Silva. Hippocampal distribution of parvalbumin neurons in female Pumpa, and Disa Jane Smee, and Ben Rattray. Exercise interventions
and male rats submitted to the same volume and intensity of aerobic for cognitive function in adults older than 50: a systematic review with
exercise. Neuroscience Letters, 2019. meta-analysis. British Journal of Sports Medicine, 2018.
[Ele19] Elena P. Moreno-Jiménez, and Miguel Flor-Garcı́a, and Julia [Kir09] Kirk I. Erickson, and Ruchika S. Prakash, and Michelle W. Voss,
Terreros-Roncal, and Alberto Rábano, and Fabio Cafini, and Noemı́ and Laura Chaddock, and Liang Hu, and Katherine S. Morris, and
Pallas-Bazarra, and Jesús Ávila, and Marı́a Llorens-Martı́n. Adult Siobhan M. White, and Thomas R. Wójcicki, and Edward McAuley,
173 174
and Arthur F. Kramer. Aerobic fitness is associated with hippocampal [Mai16] Maike Margarethe Kleemeyer, and Simone Kühn, and John Prindle,
volume in elderly humans. Hippocampus, 2009. and Nils Christian Bodammer, and Lars Brechtel, and Alexander
Garthe, and Gerd Kempermann, and Sabine Schaefer, and Ulman
[Kir11] Kirk I. Erickson, and Michelle W. Voss, and Ruchika Shaurya Prakash,
Lindenberger. Changes in fitness are associated with changes in
and Chandramallika Basak, and Amanda Szabo, and Laura Chad-
hippocampal microstructure and hippocampal volume among older
dock, and Jennifer S. Kim, and Susie Heo, and Heloisa Alves, and
adults. NeuroImage, 2016.
Siobhan M. White, and Thomas R. Wojcicki, and Emily Mailey, and
Victoria J. Vieira, and Stephen A. Martin, and Brandt D. Pence, and [Mat13] Matthew T. Schmolesky, and David L. Webb, and Rodney A. Hansen.
Jeffrey A. Woods, and Edward McAuley, and Arthur F. Kramer. Ex- The effects of aerobic exercise intensity and duration on levels of brain-
ercise training increases size of hippocampus and improves memory. derived neurotrophic factor in healthy men. Journal of Sports Science
Proceedings of the National Academy of Sciences of the United States and Medicine, 2013.
of America, 2011.
[Mat19] Matthew K. Tobin, and Kianna Musaraca, and Ahmed Disouky, and
[Kos15] Koshiro Inoue, and Masahiro Okamoto, and Junko Shibato, and Min Aashutosh Shetti, and Abdullah Bheri, and William G. Honer, and
Chul Lee, and Takashi Matsui, and Randeep Rakwal, and Hideaki Namhee Kim, and Robert J. Dawe, and David A. Bennett, and Kon-
Soya. Long-term mild, rather than intense, exercise enhances adult stantinos Arfanakis, and Orly Lazarov. Human hippocampal neuro-
hippocampal neurogenesis and greatly changes the transcriptomic genesis persists in aged adults and alzheimer’s disease patients. Cell
profile of the hippocampus. PLOS ONE, 2015. Stem Cell, 2019.
[Kri17] Krister Håkansson, and Aurélie Ledreux, and Kirk Daffner, and [Mau18] Maura Boldrini, and Camille A. Fulmore, and Alexandria N. Tartt,
Yvonne Terjestam, and Patrick Bergman, and Roger Carlsson, and and Laika R. Simeon, and Ina Pavlova, and Verica Poposka, and
Miia Kivipelto, and Bengt Winblad, and Ann-Charlotte Granholm, Gorazd B. Rosoklija, and Aleksandar Stankov, and Victoria Arango,
and Abdul K. Mohammed. Bdnf responses in healthy older persons and Andrew J. Dwork, and René Hen, and J. John Mann. Human
to 35 minutes of physical exercise, cognitive training, and mindfulness: hippocampal neurogenesis persists throughout aging. Cell Stem Cell,
Associations with working memory function. Journal of Alzheimer’s 2018.
Disease, 2017.
[Nas03] Nasser Ahmadiasl, and Hojjatallah Alaei, and Osmo Hänninen. Effect
[Kul12] Kuljeet Singh Anand, and Vikas Dhikav. Hippocampus in health and of exercise on learning, memory and levels of epinephrine in rats’
disease: An overview. Annals of Indian Academy of Neurology, 2012. hippocampus. Journal of Sports Science and Medicine, 2003.
[Lau00] Laura A. Mamounas, and C. Anthony Altar, and Mary E. Blue, and
[Nic01] Nicole C. Berchtold, and J. Patrick Kesslak, and Christian J. Pike,
David R. Kaplan, and Lino Tessarollo, and W. Ernest Lyons. Bdnf
and Paul A. Adlard, and Carl W. Cotman. Estrogen and exercise in-
promotes the regenerative sprouting, but not survival, of injured sero-
teract to regulate brain-derived neurotrophic factor mrna and protein
tonergic axons in the adult rat brain. The Journal of Neuroscience,
expression in the hippocampus. European Journal of Neuroscience,
2000.
2001.
[Lau10a] Laura Chaddock, and Kirk I. Erickson, and Ruchika Shaurya Prakash,
[Nic10] Nicole C. Berchtold, and Nicholas Castello, and Carl W. Cotman.
and Jennifer S. Kim, and Michelle W. Voss, and Matt VanPatter, and
Exercise and time-dependent benefits to learning and memory. Neu-
Matthew B. Pontifex, and Lauren B. Raine, and Alex Konkel, and
roscience, 2010.
Charles H. Hillman, and Neal J. Cohen, and Arthur F. Kramer. A
neuroimaging investigation of the association between aerobic fitness, [Ola15] Olaf Bergmann, and Kirsty L. Spalding, and Jonas Frisén. Adult
hippocampal volume, and memory performance in preadolescent chil- neurogenesis in humans. Cold Spring Harbor Perspectives in Biology,
dren. Brain Research, 2010. 2015.
[Lau10b] Laura Chaddock, and Kirk I Erickson, and Ruchika Shaurya Prakash, [Pet98] Peter S. Eriksson, and Ekaterina Perfilieva, and Thomas Björk-
and Matt VanPatter, and Michelle W. Voss, and Matthew B. Pon- Eriksson, and Ann-Marie Alborn, and Claes Nordborg, and Daniel
tifex, and Lauren B. Raine, and Charles H. Hillman, and Arthur F. A. Peterson, and Fred H. Gage. Neurogenesis in the adult human
Kramer. Basal ganglia volume is associated with aerobic fitness in hippocampus. Nature Medicine, 1998.
preadolescent children. Developmental Neuroscience, 2010.
175 176
[Rob94] Robert E. Dustman, and Rita Emmerson, and Donald Shearer. Phys- [Xia16] Xiao-Qin Wang, and Gong-Wu Wang. Effects of treadmill exercise
ical activity, age, and cognitive-neuropsychological function. Journal intensity on spatial working memory and long-term memory in rats.
of Aging and Physical Activity, 1994. Life Sciences, 2016.
[S. 95] S. A. Neeper, and F. Góauctemez-Pinilla, and J. Choi, and C. Cot- [Yaa19] Yaakov Stern, and Anna MacKay-Brandt, and Seonjoo Lee, and Paula
man. Exercise and brain neurotrophins. Nature, 1995. McKinley, and Kathleen McIntyre, and Qolamreza Razlighi, and Emil
Agarunov, and Matthew Bartels, and Richard P. Sloan. Effect of
[S. 96] S. A. Neeper, and F. Gómez-Pinilla, and J. Choi, and C. W. Cotman.
aerobic exercise on cognition in younger adults: A randomized clinical
Physical activity increases mrna for brain-derived neurotrophic factor
trial. Neurology, 2019.
and nerve growth factor in rat brain. Brain Research, 1996.
[Yan19] Yang Du, and Buyun Liu, and Yangbo Sun, and Linda G. Snetselaar,
[Sha18] Shawn F. Sorrells, and Mercedes F. Paredes, and Arantxa Cebrian-
and Robert B. Wallace, and Wei Bao. Trends in adherence to the
Silla, and Kadellyn Sandoval, and Dashi Qi, and Kevin W. Kelley,
physical activity guidelines for americans for aerobic activity and time
and David James, and Simone Mayer, and Julia Chang, and Kurtis I.
spent on sedentary behavior among us adults, 2007 to 2016. JAMA
Auguste, and Edward F. Chang, and Antonio J. Gutierrez, and Arnold
Network Open, 2019.
R. Kriegstein, and Gary W. Mathern, and Michael C. Oldham, and
Eric J. Huang, and Jose Manuel Garcia-Verdugo, and Zhengang Yang, [Yi-18] Yi-Qing Huang, and Cheng Wu, and Xiao-Fei He, and Dan Wu, and
and Arturo Alvarez-Buylla. Human hippocampal neurogenesis drops Xia He, and Feng-Yin Liang, and Guang-Yan Dai, and Zhong Pei, and
sharply in children to undetectable levels in adults. Nature, 2018. Guang-Qing Xu, and Yue Lan. Effects of voluntary wheel-running
[Sho03] Shoshanna Vaynman, and Zhe Ying, and Fernando Gomez-Pinilla. types on hippocampal neurogenesis and spatial cognition in middle-
Interplay between brain-derived neurotrophic factor and signal trans- aged mice. Frontiers in Cellular Neuroscience, 2018.
duction modulators in the regulation of the effects of exercise on
synaptic-plasticity. Neuroscience, 2003.
[Sho04] Shoshanna Vaynman, and Zhe Ying, and Fernando Gomez-Pinilla.
Hippocampal bdnf mediates the efficacy of exercise on synaptic plas-
ticity and cognition. European Journal of Neuroscience, 2004.
[Sir15] Siresha Bathina, and Undurti N. Das. Brain-derived neurotrophic
factor and its clinical implications. Archives of Medical Science, 2015.
[Sta04] Stanley J. Colcombe, and Arthur F. Kramer, and Kirk I. Erickson,
and Paige Scalf, and Edward McAuley, and Neal J. Cohen, and An-
drew Webb, and Gerry J. Jerome, and David X. Marquez, and Steriani
Elavsky. Cardiovascular fitness, cortical plasticity, and aging. Pro-
ceedings of the National Academy of Sciences of the United States of
America, 2004.
[Ste02] Stephanie Ladd Beaumont. Ocular disorders of pet mice and rats.
Veterinary Clinics of North America: Exotic Animal Practice, 2002.
[Tae18] Tae-Woon Kim, and Hye-Sang Park. Physical exercise improves cog-
nitive function by enhancing hippocampal neurogenesis and inhibiting
apoptosis in male offspring born to obese mother. Behavioural Brain
Research, 2018.
[Wil13] William D. S. Killgore, and Elizabeth A. Olson, and Mareen Weber.
Physical exercise habits correlate with gray matter volume of the hip-
pocampus in healthy adult humans. Scientific Reports, 2013.
177 178
Another crucial point that relates everything discussed above is voluntary
movement. As suggested by the term “voluntary,” these types of movements
require input from the brain. Muscles that are involved in voluntary motions
are composed of muscle cells known as muscle fibers which are controlled by
The Neural Relationship Between Smooth alpha motor neurons. Alpha motor neurons are located in the brainstem and
Movement and Musical-Motor Entrainment and spinal cord, also known as the central nervous system. These neurons and the
muscle fibers associated with the neurons are called motor units and are the
Applications in Musical Rehabilitation basis of voluntary movement. On the other hand, movements such as reflexes
occur unconsciously thus requiring no input from the brain. Rather, they work
∗ through neural circuits to rapidly and automatically carry out a motion. These
Curie Cha movements are called automatic movements. Regardless of brain input, move-
ments involve the skeletal muscles and occur as a response to either internal or
April 2, 2021 external stimuli [S. 18].
Regarding smooth movement, voluntary movement is inhibited when there
is intermittency. This unregulated and unwanted movement is seen in disorders
Abstract of movement and is called involuntary movement. Some examples of involuntary
This paper reviews the major individual areas of the nervous system re- movement are tremors and sudden jerking.
sponsible for smooth movement and neural entrainment to auditory stim- Voluntary movement is primarily controlled by the motor cortex of the brain,
uli, and how the overlapping areas are related to these two phenomena. which is located in the back of the frontal lobe. The motor cortex comprises the
The relationship between the two is then applied to suggestive methods primary motor cortex, somatosensory cortex, premotor cortex, and supplemen-
of rehabilitation. tary motor areas. The primary motor cortex contains designated regions that
are responsible for controlling certain body parts, creating a body map on the
surface of the cortex.
1 Movement There are two important descending pathways of the motor system known
as the pyramidal pathway and the extrapyramidal pathway. These pathways
Smoothness is an element of motion that is characterized by the fluidity and allow for motor signals to travel from the motor cortex to motor neurons, caus-
continuity of a movement. It can be easy to overlook this significant quality ing muscles to move. The pyramidal pathway is related to voluntary motion of
of movement and take it for granted despite its valuable purpose. Smoothness the body and the extrapyramidal pathway is responsible for automatic move-
allows for precision and predictability [T. 98]. A lack of smoothness can cause a ment. Suggested by the pyramidal pathway’s control of voluntary movement,
motion to appear jerky, which is also known as intermittency, or the fluctuation the origin of this pathway is the brain. There are two tracts that constitute the
of acceleration and deceleration in a movement [S. 15]. Hence the more intermit- pyramidal pathway: the corticospinal tract and the corticobulbar tract. The
tency, the less smooth a movement is. Many factors can influence intermittency, corticospinal tract starts at the motor cortex, specifically receiving signals from
two of the most significant being movement control and the specific task being the primary motor cortex, premotor cortex, and supplementary motor area.
performed [S. 15]. Considering these two highly variable circumstances, move- These signals travel between the thalamus and basal ganglia to the midbrain
ment smoothness can be productively studied and analyzed. It is important to and then split into two different tracts: the lateral corticospinal tract and the
distinguish smoothness from similar terms in order to effectively characterize anterior corticospinal tract. The corticobulbar tract transports motor signals to
and examine it. Similar to smoothness, the term ataxia is used to describe move the muscles of the face and neck. It starts at the motor cortex and con-
the execution of a motion. Ataxia is a condition that can be seen as a series tinues through a space between the thalamus and basal ganglia and joins with
of symptoms that cause uncoordinated movement. Ataxia can be observed in lower motor neurons to control face and neck muscles. The extrapyramidal
those with damage to critical motor areas of the nervous system and may also pathway, as mentioned before, is responsible for automatic movement. As au-
be inherited. There is an overlap between smoothness and ataxia, however they tomatic movement does not involve the cortex of the brain, the extrapyramidal
are not perfectly synonymous. Ataxia is a broader term that represents multi- pathway begins in the brain stem and travels through the spinal cord.
ple aspects of motion that may be impaired. This can include features such as
balance, oscillation, gait, posture, sensory information, and speech [T. 16]. In
essence, ataxia is an umbrella term that includes smoothness.
∗ Advised by: Andrew Savoy from the University of Chicago
179 180
Figure 2: The corticobulbar tract is shown in the diagram, beginning at the
cortex and descending towards regions of the face and neck.
Figure 1: The corticospinal tract is shown in the diagram, beginning at the
cortex and descending down the spinal cord. which stimulates the striatum. Then GABA is released from the striatum to the
external globus pallidus. The external globus pallidus, which usually hinders the
1.1 Structures activity of the subthalamic nucleus, is suppressed, so the subthalamic nucleus
has more activity. This sends excitatory signals to the internal globus pallidus,
The thalamus is a relay center that receives inputs from various areas of the brain creating more action potential and releasing more inhibitory neurons to the
and sends out signals to corresponding regions of the cortex. Concerning motor thalamus, and in turn, inhibiting movement. These two pathways work together
functions, the thalamus is composed of specific nuclei that are responsible for and balance each other to allow for coordinated movement. If this balance
movement. These nuclei, collectively known as the motor thalamus, lie near the is disrupted and there is overactivity in one of the pathways, it can lead to
motor cortex, basal ganglia, and cerebellum, all areas important to movement uncontrolled, unrestrained movement or block certain movements, depending
[5]. It is important to smooth movement because it allows signals to reach the on the pathway [H. 03]. Essentially, the coordination of the two pathways plays
important areas of the cortex. a major role in producing smooth movements. In movement disorders that cause
The basal ganglia are structures of the brain made of a network of loops and unsmooth movements such as Parkinson’s disease, there is overactivity in one
circuits that are important in voluntary movement. The basal ganglia include of the pathways. In the case of Parkinson’s disease it is the indirect pathway.
the striatum, the globus pallidus, the subthalamic nucleus, and the substantia The cerebellum is a structure near the brainstem that is important in move-
nigra. These structures are related in the two pathways of the basal ganglia: ment and learning. It can be divided into three functional regions: the cere-
the direct pathway and the indirect pathway. These pathways allow wanted brocerebellum, the spinocerebellum, and the vestibulocerebellum. The cerebro-
movements to occur and inhibit unwanted movements. The direct pathway cerebellum is related to movement planning and learning. The spinocerebellum
excites the thalamus and allows it to send signals to other areas to produce receives sensory data from the spinal cord and information about limb loca-
movement. It starts at the cortex and synapses with neurons in the striatum, tion. The vestibulocerebellum controls balance and posture. In relation to
releasing the neurotransmitter glutamate that excites the inhibitory neurons of smooth movement, the cerebellum allows for the prediction and correction of
the striatum. GABA is released and inhibits the internal globus pallidus and motor functions. This is evident in the effects of cerebellar damage. Those with
substantia nigra. Under no change, the thalamus is usually inhibited by the cerebellar damage can experience motor deficits such as ataxia, tremors, and
globus pallidus but by inhibiting the inhibitor, the thalamus is not subdued, as it difficulties with aim; these can result in unsmooth movements. This suggests
normally is, but excited and more activity occurs. This sends more signals from the cerebellum plays a necessary role in motor execution and planning [A. 06].
the thalamus to the cortex then eventually to the muscles to produce movement.
On the other hand, the indirect pathway inhibits movement. It starts at the
cortex and synapses with the neurons of the striatum, releasing glutamate,
181 182
Figure 3: This is a feedback pathway diagram of the basal ganglia and the right
shows how the pathways are connecting the areas of the brain.
Figure 4: The ascending auditory pathway involving the cochlear nuclei, supe-
2 Entrainment rior olive, inferior colliculus, and medial geniculate nucleus.
Entrainment can be described as the locking of frequencies between two sepa-

many of the studies proved the cerebellum’s involvement in auditory processing.
rate systems. This means the occurrence of one event lines up with the occur-
So along with the auditory pathway, the cerebellum can play a role in music
rence of another. A common example is the synchronization of metronomes ini-
processing, therefore entrainment as well.
tially clicking at different times. Eventually the clicking becomes synchronized
The inferior colliculus is a part of the auditory pathway and is crucial in
through a common medium such as a table the metronomes are placed on. In
entrainment. The IC sends information to the thalamus which is sent to the
the instance of music, the brain entrains to the beat of the music. When this
cortex but it also sends signals to the cerebellum; it receives input from auditory
occurs, the firing rates of auditory neurons and motor neurons are entrained [M.
structures and cerebral cortex and sends information to dorsolateral pontine
15], therefore creating a period of rhythmic synchrony. This involves the audi-
nuclei [A. 13]. This structure is crucial because it is a component that is early
tory system and calls attention to the high connectivity between the auditory
in the auditory pathway and has projections to motor-related areas.
and motor systems. Important structures involved in rhythmic entertainment
are the auditory pathway, cerebellum, and inferior colliculus.
It is important to discuss the auditory pathway regarding entrainment to 3 Synthesis
music. The auditory pathway allows sound to be taken in and processed by the
brain. This pathway starts with the cochlear nucleus; this structure receives After discussing the areas of the nervous system related to movement and neural
input from the inner ear and sends information to the superior olive and the entrainment to auditory input, it is evident there is an overlap of common
lateral lemniscus. Then, information is sent to the inferior colliculus, which is a structures within the two domains. These areas could be the answer to effective
key structure in entrainment. Afterwards, signals arrive at the medial geniculate treatment for movement disorders.
body, which is in the thalamus. The thalamus relays information to multiple The cerebellum seems to be directly involved with both activities of move-
areas of the brain and in this case, mainly the auditory cortex. At the auditory ment and entrainment. However, it is important to further examine the high
cortex the signals are projected across many areas of the brain that allow the connectivities of the cerebellum with other structures. These neural connections
sound to be processed. will surely be affected in various motor deficits and therefore the treatment of
those motor deficits; for instance, the cerebellum and basal ganglia. As the
2.1 Structures cerebellum and basal ganglia have connections through the dorsolateral pon-
tine nuclei, motor deficits resulting in damage to one of those two areas can be
The cerebellum is involved with entrainment, specifically with sensorimotor syn- detrimental, especially in overlapping functions. Further, major areas influenc-
chronization and the auditory system. The cerebellum is associated with sensory ing movement can be affected due to the circuitry and connectivity of movement
information and this includes auditory stimuli. A meta-analysis showed that ar- execution.
eas of the cerebellum were active during auditory stimuli [A. 05]. This means in The link between the two main ideas of movement smoothness and entrain-
183 184
ment becomes relevant when discussing methods of neurorehabilitation. Due [J. 07] J. M. Hausdorff, J. Lowenthal, T. Herman, L. Gruendlinger, C. Peretz,
to the predictive nature of rhythmic entrainment, it can provide a pattern that and N. Giladi. Rhythmic auditory stimulation modulates gait variability
outlines the time for which a movement must be executed [8]; this is the main in parkinson’s disease: Effects of ras on gait variability in pd. European
concept. The structure music can lay out partly compensates for the lacking Journal of Neuroscience, 2007.
predictive model.
There have been many different uses of music in therapies. A method that [M. 14] M. A. G. Witek, E. F. Clarke, M. Wallentin, M. L. Kringelbach, and
uses rhythmic stimulation is known as Rhythmic Auditory Stimulation or RAS. P. Vuust. Syncopation, body-movement and pleasure in groove music.
It uses metronomic beats in music to provide timing for walking. RAS is com- PLOS One, 2014.
monly used for Parkinson’s patients in gait rehabilitation. This method of music [M. 15] M. H. Thaut, G. C. McIntosh, and V. Hoemberg. Neurobiological foun-
therapy and similar methods applies the neurobiology explained. RAS has been dations of neurologic music therapy: rhythmic entrainment and the
proven to be successful for these patients evident in stride length and speed [J. motor system. Frontiers in Psychology, 2015.
07]. This relates to the pattern or time frame that is established by rhythm
and music. This can be practiced further in future therapeutic techniques. Per- [S. 15] S. Balasubramanian, A. Melendez-Calderon, A. Roby-Brami, and E.
haps the application of musical melody compared to just rhythm can affect the Burdet. On the analysis of movement smoothness. Journal of Neuro-
outcome. Maybe even the syncopation of rhythm can yield different results. Engineering and Rehabilitation, 2015.
Syncopation is an aspect of music that is characteristic of off-beat rhythms and
differing beats. This can evoke desire to move and this approach may be more [S. 18] S. Armstrong, M. V. Sale, and R. Cunnington. Neural oscillations and
influenced by affective means as well [M. 14]. In culmination, crucial regions the initiation of voluntary movement. Frontiers in Psychology, 2018.
of the brain, body, and auditory system work together to allow functions of [T. 98] T. J. Sejnowski. Making smooth moves. Nature, 1998.
movement and entrainment. When movement is inhibited or uncontrolled in
some way, these motor areas are lacking proper function. With the absence of [T. 16] T. Ashizawa and G. Xia. Ataxia. Continuum Lifelong Learning in
proper function, there comes a need for external stimuli to promote rehabilita- Neurology, 2016.
tion of the impacted structures. This is where entrainment and music therapy
can play a crucial role in motor rehabilitation. The strengthening connections
and unique aspects between entrainment and movement can target the problem
in a different way that may not be possible with other means.
References
[A. 05] A. Petacchi, A. R. Laird, P. T. Fox, and J. M. Bower. Cerebellum
and auditory function: An ale meta-analysis of functional neuroimaging
studies. Human Brain Mapping, 2005.
[A. 06] A. J. Bastian. Learning to predict the future: the cerebellum adapts
feedforward movement control. Current Opinion in Neurobiology, 2006.
[A. 13] A. Tierney and N. Kraus. The ability to move to a beat is linked to
the consistency of neural responses to sound. Journal of Neuroscience,
2013.
[C. 13] C. Bosch-Bouju, B. I. Hyland, and L. C. Parr-Brownlie. Motor tha-

lamus integration of cortical, cerebellar and basal ganglia information:
implications for normal and parkinsonian conditions. Frontiers in Com-
pututational Neuroscience, 2013.
[H. 03] H. J. Groenewegeni. The basal ganglia and motor control. 2003.
185 186
A NATO intervention: mandate of civilian protection without affecting regime change? NATO did not
have an ulterior motive of removing Gaddafi, instead, they actually could not
complete their aim of civilian protection with Gaddafi in power. This does not
The Right Choice for Libya mean that they took any direct aims at Gaddafi’s government, rather they fought
for civilian protection and in turn, Gaddafi’s forces died as collateral damage. In
addition, if we examine the boundaries of ‘R2P’ and UN R-1973 we see that a
regime change is not out of the question and it is even considered in UN R-1973.
Anya Nedungadi Therefore, the exclamations of the African Union and Russia that declared that
NATO was overstepping its instructions were misinformed because the
international community had accepted the possibility of regime change before the
intervention even began.
1 、 T HE NATO INTERVENTION HAD THE INTENTION OF REGIME The UN Resolution 1973 did not specifically mandate a regime change but
CHANGE allowed ‘all necessary measures’ as well as finding a ‘solution…which responds
to the legitimate demands of the Libyan people…to lead the political reforms
necessary to find a peaceful and sustainable solution.’ 7 Interpreting this text
NATO arguably intervened with the ‘wrong intent’; they pushed for a regime delivers its own set of challenges; it might either allow permissive or restrictive
change instead of focusing on saving civilian lives. During the war, when NATO actions depending upon one’s political and moral perspective.8 If NATO could help
exhibited questionable actions, scholars believed that NATO was not operating the Libyan people without overthrowing Gaddafi, this could not be condemned in
within the boundaries of the mandate ‘Responsibility to Protect’ that the the eyes of UN Resolution 1973. In reviewing statements made by coalition
international community (led by the UNSC) agreed that NATO would intervene members following the UN Resolution 1973, the Libya Contact group, members
under. The UN mandate ‘Responsibility to Protect’ is defined by the United of the London conference and North America concluded that it would be
Nations Office of Genocide Prevention and Responsibility to Protect as an impossible to fulfil the ‘Responsibility to Protect’ mandate with Gaddafi in power,
international norm that seeks to ensure that the international community does not particularly considering his government’s deliberate denial of electricity, water,
fail to halt the crimes of genocide, war crimes, ethnic cleansing and crimes against fuel and food to ordinary Libyans. 9 This means that Gaddafi’s death and the
humanity.1 It does not mandate a regime change. However, the war that followed subsequent power vacuum could not be blamed on NATO both because the UN
in Libya was one in which the Western powers worked in close collaboration with mandate under which they intervened and included the possibility of regime change,
the rebel forces; serving them with their air forces, but also providing them with and because there was an international consensus that a regime change was an
arms, training and propaganda support.2 The NATO powers also supposedly had expected occurrence. NATO cannot and should not be penalised for operating
hundreds of operatives on the ground in Libya, training the rebels and giving them appropriately under a legitimate authority.
intelligence and other support, therefore violating UN resolution 1973’s
prohibition of an occupation force “in any form” and overstepping their initial aim It was impossible to meet the mandate of civilian protection without a regime
of a no-fly zone.3 On the day that the aerial attacks on Libya started, concerns were change. The intervention’s chance of success of lowering civilian casualties by
already raised about military overreach. The Chinese government expressed regret protecting civilian areas should not be inhibited by trying to maintain the current
at the American and European assault on Libya, and Russia condemned the attack.4 regime. In response to Alan J Kuperman, this paper argues that NATO did actually
The African Union began stressing that only dialogue and consultation could bring have a very limited scope of intervention which supports its sole aim of civilian
solutions in Libya.5 Scholars (Gregory, 2016)6 declare that the Libyan intervention protection. They intervened with a no-fly zone which means that they didn’t have
was no different to previous invasions in the Middle East because NATO’s actions troops on the ground and they only bombed areas where Gaddafi was harming
suggested that they were more focused on killing Gaddafi rather than saving civilians.10 For example, at the time NATO and its allies initiated airborne attacks
civilian lives. Therefore, a NATO intervention was the wrong choice for Libya. on pro-Gaddafi forces in the city of Misrata, the city had already been under attack
In response to scholars who believe that NATO intervention had an ulterior by government tanks and artillery for several days.11 This was also the case when
motive (Forte, 2012), I would like to pose a question: Was it possible to meet the NATO began its airstrikes of Gaddafi’s troops within Ajdabiua, where government
soldiers, tanks and warplanes had been bombarding the town.12 In considering that
NATO dropped about 8,000 bombs in almost 18,000 sorties, the results were
1
United Nations Office of Genocide Prevention and Responsibility to Protect impressive: there were few civilian deaths. This success was due to the precision-
<https://www.un.org/en/genocideprevention/>
2

ibid
3 7
Forte, Maximilian Christian 2012. Wedgwood, Andrew, and A. Walter Dorn (2015)
4 8
The Guardian, (2011) ‘Libya Attacks under Way’. ibid
5 9
African Union, press release (2011) ‘The African Union Deeply Concerned about the ibid
10
Situation in Pattison, James (2011)
11
Libya’ Ulfstein, Geir, and Hege Fosund Christiansen (2013)
6 12
Gregory, Robert H (2015) ibid.
187 188
guided munitions and meticulous planning based on solid intelligence for each Forbes article,18 when compared to dangerously infamous dictators such as Stalin,
NATO action. 13 It was impossible to escape this war without casualties on we see that the two leaders follow the same principles;
Gaddafi’s side because they were the ones committing atrocities and murder,
however, the precision by which NATO handled their intervention proved that they 1) No alternative source of power or authority, no matter how seemingly weak
were not intent on killing any more of Gaddafi’s forces than what was needed. All or trivial.
of these precautions reinforce the fact that NATO intervened with the intent of
civilian protection, and any other political or social ramifications were only
2) No freedom of the press or expression. Even private derogatory comments
collateral damage.
are criminal acts.
2 、 Gaddafi’s actions were not serious enough to warrant a military 3) Imprison, execute or banish all enemies, actual or potential.
intervention
4) Do not worry about punishing innocent people.
Critics (Pattison, 2011) of the NATO intervention assert that the Western
media exaggerated the actions of Gaddafi in order to portray him as the main 5) At all costs, prevent any combination of internal and external enemies.
perpetrator of mass atrocities, consequently, the situation in Libya did not need an Failing to do so can be fatal.19
intervention. In Eastern Libya, where the uprising began as a mix of peaceful and
violent protests, Human Rights Watch documented only 233 deaths in the first days Gaddafi was a classic, unconstrained dictator; he followed similar ideology
of the fighting, not 10,000, as had been reported by the Saudi news channel Al to Stalin and he was not against murdering innocent people for power. Therefore,
Arabiya.14 The rebels had put up very little defence against government forces; when critics state that the conflict was on the ‘verge of ending’ before NATO
with their rudimentary equipment and inadequate training, they were in no shape intervened, I am surprised that they did not view him as a future threat. The 1988
to win a war. Therefore, Gaddafi’s forces would have perhaps captured Benghazi bombing of Pan Am Flight 103 over Lockerbie, Scotland, the 1989 Union de
by March 20th, thereby supposedly ending the one-month conflict at a total cost of Transports Aériens [UTA] Flight 772 bombing, and the 1996 massacre of 1,000
just over 1,000 lives. On March 17th, Gaddafi pledged to protect the civilians of inmates at the Abu Salim prison were crimes directly connected to the Gaddafi
Benghazi, as he had those with other recaptured cities, adding that his forces had regime.20 Internally, Gaddafi oppressed Libyans through the banning of foreign
‘left the way open’ for the rebels to retreat to Egypt.15 The violence was supposedly language education, restricting travel, eliminating political parties, criminalising
on the ‘verge of ending’ when NATO decided to supply rebel forces with minority cultures, and developing a variety of oppressive internal security
ammunition, giving them the strength that they needed to carry on the war. organisations.21 These actions show Gaddafi’s disregard for his peoples’ lives and
his unwillingness to protect his own citizens.
Critics also claim that there were many occurrences of misinformation where
the Western media exaggerated attacks by Gaddafi’s government. Supposedly the
In addition, a report released by the UNHRC concluded that Gaddafi’s forces
number of deaths was minimal-- only 24 protestors in three days were killed. 16
had committed a wide range of abuses against his people such as rape, torture,
According to Human Rights Watch, the death count of protestors in Libya was less
murder, indiscriminate attacks and a range of other human rights abuses. If there
than the number of black mercenaries executed by the Libyan rebels in mid-
remained any doubt about Gaddafi’s potential for future human rights abuses,
February and fewer than the number of protester deaths in Tunis or Egypt which
Gaddafi erased this doubt himself by claiming on the 17th March 2011, ‘We will
elicited no intervention or UNSC aid.17
come house by house, room by room… We will find you in your closets. We will
Gaddafi’s actions did not need a military intervention; the number of have no mercy and no pity.’ Gaddafi wasn’t content with just recapturing a city, he
causalities were low despite the Western media’s exaggerations and the rebel wanted justice and he wanted to ‘cleanse Libya,’22 actions that would have resulted
forces were on the verge of being subdued. in more mass atrocities against civilians.
While this argument has some validity, it fails to recognise that there was no
way to prevent future mass atrocities from being committed without an intervention Finally, the NATO intervention was needed to combat Gaddafi’s forces
by NATO. Gaddafi was a ruthless, unpredictable dictator, desperate to keep his because it was clear that he was becoming a potential threat not only to domestic
power without consideration for the deaths of innocent people. As shown in a 2011 security, but to international security. By ignoring the 1970 UNSCR resolution,
which demanded an arms embargo, a travel ban and an assets freeze, he
demonstrated his inability to comply with non-military intervention and his
apparent disregard of the international community’s decisions. The Security

13 18
Pattison, James (2011) Gregory, Paul (2011)
14 19
Wedgwood, Andrew, and A. Walter Dorn (2015) ibid
15 20
Kuperman, Alan J (2013) Wedgwood, Andrew, and A. Walter Dorn (2015)
16 21
Forte, Maximilian Christian 2012. ibid
17 22
ibid Morris, Justin (2013)
189 190
Council only then took measures to enforce a military intervention. Again and NATO cannot be blamed for the Libyan civil war which reinforces my
again, Gaddafi showed that he was capable of massacres and he was unafraid of argument that the intervention was the right choice.
the consequences. The Libyan situation was unlike previous Western interventions
in the Middle East. The intervention was not based on dodge dossiers of faked 4、The comparison of Libya and Syria
evidence and lies, but on the live unfolding of crimes against humanity. Gaddafi
was a potential danger to the lives of the Libyans as well as to international security,
therefore the NATO intervention was a necessity. The Syrian civil war is a very beneficial case study as it proves a parallel for
scholars to draw on between the road of war with a military intervention, and the
road of war without. Syria did not receive the help of a NATO intervention because
3、Introduction to the argument in support of the NATO intervention NATO’s chosen means of implementing the mandate ‘R2P’ in Libya paved the
path for those who were sceptical of the mandate to delegitimize the norm. This
As the 8-year-long Libyan civil war continues to wreck the lives of ordinary resulted in the prevention of international aid to Syria, where it was needed the
Libyan civilians, NATO has been blamed on multiple accounts of starting the most.
political instability in Libya. The country has been riddled with insecurity since the
removal of Gaddafi: multiple factions are vying for control particularly within the Both countries were at war due to the “Arab Spring” uprisings of 2011 which
capital.23 With competing governments in the country’s east and west, and armed triggered a wide set of social movements and regime change across the Middle East
militias, who are not held accountable for any of their actions, controlling large and North Africa.27 The two countries had similar sizes of protester deaths in 2011-
parts of the country and exerting coercive political influence, Libya is effectively -400 for Syria, and at least 233 for Libya, according to Human Rights Watch,28 and
unable to provide for its people.24 In exploring how Libya got to this point, analysts, were both being ruled by ruthless dictators who were abusing their citizens.
journalists and politicians often point their fingers at the 2011 intervention. Many However, the death count in Syria is vastly greater than the death count in Libya.
have said that the mess they are in is what inevitably ensues when NATO Since 2011, the Syrian civil war has lowered life expectancy by 20 years. Syrian
intervenes in a country. But who should actually be blamed for the outbreak of civil civilians have constituted 70.6% of deaths (101,435) compared with the 29.4% of
war? deaths of oppositionist combats (42,177). Proportions of children among civilian
It was failures of the international community after the intervention that deaths increased from 8.9% (388 of 4254 civilian deaths) in 2011 to 19.0% (4,927
resulted in a civil war. The NATO intervention was extremely successful in of 25,972) in 2013 and to 23.3% (2,662 of 11,444) in 2016.29 In Libya, there was
preventing the immediate massacre of civilians, and if it wasn’t for their actions, an estimated minimum of only 727 total civilian deaths.30 Yes, proportionally Syria
the beloved city of Benghazi would most likely not exist today. As the British is a larger country with 22.5 million people versus Libya’s 6.6 million people31
Prime Minster David Cameron put it: doing nothing in Libya would have been to which may account for a slight variation of death count, but considering that these
‘facilitate murder.’ 25 NATO succeeded in its job to prevent this murder and wars started out so similarly, the difference in death count is shocking.
followed through in all areas of the resolution that they intervened under. It was
the failure of the international community, specifically the UK, to build a strong A NATO intervention was ultimately the right choice for Libyan citizens
and stable state after the intervention that caused a civil war to break out in Libya. because it prevented the spread of wide scale terrorism and death as it is in Syria-
A 2016 UK foreign affairs committee report criticised David Cameron, the British where Libya has as many as 6,000 Islamic State soldiers on the ground, Syria has
Prime Minster, for failing to develop a plan for post-intervention Libya, pointing 20,000 to 30,000 according to the CIA- and because politically, Libya is faring
out that the amount of funds spent on development - £25m ($31m) - was less than much better than Syria. After a long negotiation led by the U.N., a Government of
a tenth of the cost of the actual intervention.26 “Your friends in Britain and France National Accord is working to consolidate power in Libya;32 success is not assured,
will stand with you as you build your country and build your democracy for the but there is a clear path forward, a growing consensus around that path, and a
future,” Cameron promised - but in practice, that support was in short supply, reasonable chance of real political reconciliation. This is very unlike the chaos that
particularly as time wore on and attention drifted elsewhere. Ironically, those who is occurring in Syria where amidst the chaos of fighting between the government
blame the 2011 intervention for the crisis in Libya are guilty of precisely the same and anti-government fighters, the Islamic State is taking over large parts of Iraq
mistake as the intervention itself: not paying any further attention to the country and then moving into eastern Syria, where they are gaining land and power.33 A
after September 2011, until it had fallen apart. They accuse advocates of
intervention of seeing it as the solution to every problem, but they themselves
disregard all nuance, advocating avoidance as the solution, irrespective of the
problem.
27
Shapiro, Ari (2013) ‘Why Syria is more complicated than Libya”
28
Human Rights Watch <https://www.hrw.org>
29
Guha-Sapir, Debarati (2018)
30
Washington Post (2020) “Civilian casualties surge in Libya during Tripoli battle, study
23
Green Matthew (2019) finds”
24 31
Dahan, Nadine (2019) The Guardian (2011) “How Libya and Syria compare”
25 32
ibid Rogin, Josh (2016) ‘Obama’s Biggest mistake isn’t Libya. Its Syria’
26 33
ibid https://www.bbc.co.uk/newsround/16979186
191 192
NATO intervention was the right choice for Libya because it prevented the
disasters that have occurred in Syria.
5、Conclusion Wedgwood, Andrew, and A. Walter Dorn. "NATO’s Libya Campaign 2011:
Just or Unjust to What Degree?" Diplomacy & Statecraft 26.2 (2015): 341-362.
A NATO intervention was the right choice for Libya in 2011. In response to
this conclusion, we must rethink the delegitimization of the norm R2P and accept
the mandate as a suitable and successful means of military intervention. As proven
in this paper, an intervention under this mandate can prevent significant life loss Forte, Maximilian Christian. Slouching Towards Sirte: NATO's War on
and can protect civilians, if carried out in the appropriate manner. If the Libya and Africa. Montreal: Baraka books, 2012.
international community hadn’t been so eager to rule out the norm after the NATO
intervention in Libya, many lives in other Middle Eastern countries, such as Syria,
could have been saved. The information gathered about the use of the mandate
during the intervention in Libya should be used to strengthen the boundaries and
Pattison, James. "The ethics of humanitarian intervention in Libya." Ethics &
analyse the methods of military intervention so that it continues to save lives.
International Affairs 25.3 (2011): 271-277
Morris, Justin. "Libya and Syria: R2P and the spectre of the swinging
pendulum." International Affairs 89.5 (2013): 1265-1283.
Bibliography Guha-Sapir, Debarati, et al. "Patterns of civilian and child deaths due to war-
related violence in Syria: a comparative analysis from the Violation
Documentation Center dataset, 2011–16." The Lancet Global Health 6.1 (2018):
Kuperman, Alan J. "Obama's Libya debacle: how a well-meaning e103-e110.
intervention ended in failure." Foreign Affairs 94.2 (2015): 66-77.
Karniel, Yuval, Amit Lavie-Dinur, and Tal Azran. "Broadcast coverage of

Nuruzzaman, Mohammed. "The 'Responsibility to Protect' Doctrine: Revived Gaddafi's final hours in images and headlines: A brutal lynch or the desired death
in Libya, Buried in Syria." Insight Turkey 15.2 (2013): 57. of a terrorist?" International Communication Gazette 77.2 (2015): 171-188.
Kuperman, Alan J. "A model humanitarian intervention? Reassessing Ulfstein, Geir, and Hege Fosund Christiansen. "The legality of the NATO
NATO's Libya campaign." International Security 38.1 (2013): 105-136. bombing in Libya." Int'l & Comp. LQ 62 (2013): 159.
Averre, Derek, and Lance Davies. "Russia, humanitarian intervention and the Levitt, Matthew A. "The political economy of Middle East terrorism." Middle
Responsibility to Protect: the case of Syria." International Affairs 91.4 (2015): 813- East Review of International Affairs 6.4 (2002): 49-65.
834.
Odeyemi, Christo. "R2P intervention, BRICS countries, and the no-fly zone
Carpenter, Ted Galen. "Tangled web: The Syrian civil war and its measure in Libya." Cogent Social Sciences 2.1 (2016): 1250330.
implications." Mediterranean Quarterly 24.1 (2013): 1-11.
193 194
Dahan, Nadine. “Libya’s ongoing crisis shouldn’t be blamed on the 2011
intervention” Available from <https://www.middleeasteye.net/opinion/dont- What, if Anything,
blame-2011-intervention-libyas-ongoing-crisis>
is Wrong with Surveillance Capitalism?
Human rights Watch “Unacknowledged deaths: Civilian Casualties in
NATO’s air campaign in Libya” Available
<https://www.hrw.org/report/2012/05/13/unacknowledged-deaths/civilian-
from Aaron Chen
casualties-natos-air-campaign-libya>
Abstract
Gregory, Paul “Colonel Gaddafi’s lesson for dictators” Available from
This paper tackled existing literatures and evidences on the emergence of a
<https://www.forbes.com/sites/paulroderickgregory/2011/10/30/colonel-gaddafis-
supposedly new socioeconomic phenomenon —"surveillance capitalism”—
lesson-for-dictators/>
focusing specifically on the question “what, if anything, is wrong with
surveillance capitalism?” I explored this by referencing three established
The Guardian “Libya Attacks under way- Saturday 19th March part 2” conceptions of freedom: freedom as autonomy, non-interference, and non-
Available from <https://www.theguardian.com/world/blog/2011/mar/19/libya-no- domination. On autonomy, I explored how the consumerist landscape fueled by
fly-zone-live-updates> surveillance capitalism may encroach upon the authenticity and rational capacity
of consumers, thus preventing individuals from leading a “good” life; On non-
interference, I analyzed the implications of data’s implementation, considering
Gregory, Robert H. Clean bombs and dirty wars: air power in Kosovo and the nuances behind its support; On freedom as non-domination, I shifted away
Libya. U of Nebraska Press, 2015. from applications of data, and instead, evaluated how the act of appropriating data
itself may be problematic. In considering the unprecedented arbitrary power that
firms are able to accumulate during the appropriation-stage, I have shown how
Shapiro, Ari (2013) ‘Why Syria is more complicated than Libya” Available surveillance capitalism manifests in ways ethically unacceptable, intrinsic, unique,
from <https://www.npr.org/2013/08/29/216858049/why-syria-is-more- and dangerous in social implications.
complicated-than-libya>
Rogin, Josh (2016) “Obama’s Biggest mistake isn’t Libya. Its Syria” 1、INTRODUCTION
Available from
https://www.bloomberg.com/opinion/articles/2016-04-11/obama-s-biggest- Just as the abundance of human labor once was to the foundation of industrial capitalism, it seems
mistake-isn-t-libya-it-s-syria that a new information civilization characterized by its ubiquitous digital traces has similarly
birthed the emergence of what various contemporary scholars deem to be "surveillance
capitalism", an economic order founded upon the extraction of this new data. This paper, though,
Hobson, Christopher. "Responding to failure: The responsibility to protect is not so much concerned with a historical account that traces how surveillance capitalism came
after Libya." Millennium 44.3 (2016): 433-454. to develop. Rather, it will focus on a critique of this new economic order: what, if anything, is
wrong with surveillance capitalism?
The Guardian (2011) “How Libya and Syria compare” Available from <
https://www.theguardian.com/world/2011/apr/28/syria-libya-how-they-compare> Before moving on, it is important to establish what is meant by ‘wrong’. An economic order such
as surveillance capitalism can be wrong in various ways— it can be dysfunctional (a functional
critique), where it fails to function as intended. For example, a knife used for cutting is said to be
Washington Post (2020) “Civilian casualties surge in Libya during Tripoli battle, dysfunctional if it has a blunt blade1. The cutting knife can also be inherently dysfunctional if it
study finds” Available from <https://www.washingtonpost.com/national- does not have a blade or is equipped to function in a way that is actively contradictory to what is
security/civilian-casualties-surge-in-libya-during-tripoli-battle-study- intended2; an economic order can also be morally abhorrent (a moral critique), where the system
finds/2020/06/01/17e4d7a2-a408-11ea-8681-7d471bf20207_story.html> is unfair and exacerbates injustice; a system can also be ethically unacceptable (an ethical
1
Rahel Jaeggi, “What (If Anything) Is Wrong with Capitalism? Dysfunctionality, Exploitation and
Alienation: Three Approaches to the Critique of Capitalism.” pp. 44-65.
2
In the case of capitalism, if one assumes that such a system is supposed to function by allocating
resources to fulfill consumer needs, then it can be inherently dysfunctional if it indeed prevents resource
195
allocation or prevents the fulfillment of those needs. In this scenario, capitalism functions against its
purpose.
196
critique3), where it inhibits us somehow from leading and living a fulfilling life. This paper is 4. whether this 'wrong' is worse than previous forms of capitalism in its effects,
only interested in the ethical critique of surveillance capitalism, as it pertains to a reasonably which will constitute a general discussion of the economic and social impact that results
broad exploration of implications that such a system has on our mode of living and interacting, from some feature of surveillance capitalism. Both economic and social consequences
and thus leaves ample room for normative judgment and debate. are relevant, given that surveillance capitalism — a new form of capitalism — is at the
end of the day an economic system that has close ties to the public sphere.
In order to further focus this paper, surveillance capitalism will only be assessed from the lens of
freedom and the various conceptions of it, namely freedom as autonomy, non-interference, and If these latter two prongs are met, then providing that there is an identifiable wrong in surveillance
non-domination. Surveillance capitalism’s implications on the preservation of freedom is quite a capitalism, I can claim that 'wrong' manifests in a unique and worse manner in surveillance
relevant and potent topic, to be sure, as freedom is often regarded as a prerequisite to leading a capitalism in comparison to alternative forms of capitalism.
fulfilling life—a popular standard by which to gauge whether a system is ethically acceptable.
At this point, then, one can begin to develop a normative criterion for assessing surveillance 2、WHAT IS SURVEILLANCE CAPITALISM?
capitalism: To be considered ‘wrong’, some implication that the system has on our freedom must
be Before moving on to the normative concerns of surveillance capitalism, it is important to
understand its underlying economic logic. Essentially, what is surveillance capitalism?
1. ethically unacceptable, insofar as it inhibits us somehow from leading and
living a fulfilling and happy life Surveillance capitalism is interpreted as an economic logic that emerges with the practice of data
extraction, in which it "claims human experiences as free raw material" for private appropriation
It will also be added that what can be identified as ethically unacceptable must be and commercial practices. 4 These "human experiences as free raw material" become what is
understood in this paper as 'data'. Interpreted in such a manner, then, we must first understand the
2. intrinsic to the surveillance system, where surveillance capitalism as it is cannot nature of data itself before moving on to an exploration of surveillance capitalism as a whole.
be conceived without the feature that perpetrates the identified ‘wrong’.
First, data under surveillance capitalism is extracted at an unprecedented “scale”.5 Being 'free',
This second prong is imperative, as to identify a feature that is only wrongful in some instances data under surveillance capitalism differs from other forms of assets in that the notion of 'finite
is only telling to the extent that surveillance capitalism could be implemented in a way that has resource' that applies to all other factors of production does not apply. While what it is used for
perverse implications, not that it is inherently concerning-- which is not enough for the purposes can have value, it itself has no inherent cost. In combination with better technological
of this paper. advancements that allow much of the extraction process to be undergone in an opaque manner in
which consumers and regulators are largely ignorant of, this transforms the economic logic of
If the two aforesaid requirements are met, then there must be something that is wrong with production by allowing firms to extract data on an unprecedented scale in an ever-going manner.
surveillance capitalism. That being said, if this paper fails to meet these requirements, conclusions
only stem so far as the ethical critique of surveillance capitalism; perhaps, one might resolve to Second, being "human experiences", data now profiles consumers and identifies them with
different conclusions when assessing surveillance capitalism from other perspectives. "scope" and "depth".6 It can range from a consumer's digital footprint, to behavioral patterns, and
to even offline information pertaining to one’s socioeconomic demographic.7 What is crucial to
However, as it currently is, this metric is quite trite. Surely, it wouldn't be quite so meaningful if note is that consumers don't need to actively provide any of this data—as opposed to limited
what we find evidently wrong in surveillance capitalism also be happens to manifest in other information actively given up by the consumer in traditional forms of data collection. For
forms of capitalism, economic systems, or human institutions. If that is the case, we can at best example, as opposed to optional surveys and rating features that are traditionally used to gauge
claim disillusionment against the generic modern society and human civilization, which isn’t consumer interest and preferences online, we are seeing a surge in the use of browser cookies,
quite sufficient for the purpose of this paper. Or worse, if whatever wrong we identify is where companies place cookies on their websites and pay for permission to place cookies across
comparatively worse under other, for example, previous manifestations of capitalism in terms of the internet so as to effectively relay data pertaining to user identification and user activity back
social and economic impacts, that would imply surveillance capitalism to be in fact an to websites—all of which is done via technology and without necessarily requiring a user's active
improvement for the better. input.8
Hence, I will add two potential points of discussion on top of the aforesaid metric that primarily
focuses on the comparatives to surveillance capitalism: 4
Shoshana Zuboff, The Age of Surveillance Capitalism: The Fight for the Future at the New Frontier of
Power.
3. whether the 'wrong’ identified is unique to surveillance capitalism in 5
Ibid.
comparison to alternative forms of capitalism
6
Ibid.
7
Karen Yeung, “Five Fears about Mass Predictive Personalization in an Age of Surveillance Capitalism.”
pp. 258-269.
3
Here, the differentiation between ethical and moral critique refers to a common philosophical distinction
8
for the sake of discussion. See Richard Kraut, “Aristotles Ethics”.; for a more specific account pertaining Sarah Myers West, “Data Capitalism: Redefining the Logics of Surveillance and Privacy.” pp. 20-41.
to capitalism, see also Ibid.
197 198
These characteristics are intimately related to data's new functions, where because of data’s the means of production, wealth and power is at the very least disseminated amongst a class of
“scale”, “scope”, and “depth” in profiling consumers, data no longer just tells what consumers individuals involved in different corporate bodies; this ensures a relative degree of competition
explicitly want, but can tell firms who consumers are as decision-making individuals beyond and interdependence between stakeholders within free market economies. Under surveillance
what consumers are willing to reveal. Essentially, while traditional forms of data may tell firms capitalism, however, since the extreme "concentrations of data, wealth, and power” is
'what' consumers want, they don't tell firms 'why' — yet now data does. Consequently, unlike monopolized by a controlled handful of firms,
data's traditional applications, in which information is compiled for the improvement of services
and better fulfillment of consumer needs, data in its new form is collected so that firms can also 3.) this “concentration” leads to the establishment of total market control on an
interpret how consumer choice is formed on an individual basis with optimum accuracy and unprecedented level.
precision—"total certainty”. 9 Data's value is thus underpinned by not just telling what the
consumer wants and how to meet those wants, but also how to make a particular consumer want First, this degree of market control is unprecedented because it is expansive in nature and not
a product irrespective of its actual utility. limited to a specific industry: For example, unlike a monopolist that develops certain agricultural
technologies specific to its industry, surveillance capitalists’ authority (such as Google’s) is
Given that these changes to data’s characteristics and functions are driven by technological recognized in across sectors. This is powerful because the economic value of data extraction and
development, data and its extraction in this new form is a consumer profiles in aiding profit maximization is something that can be recognized as important
by most profit-driven firms; the vastly favorable reception towards the surveillance capitalist
3.) unique phenomenon to information civilization and the surveillance capitalism that logic makes it possible for data extraction techniques to be applied and reproduced in many
emerged with it. sectors. This allows established surveillance capitalists to replicate their wealth, barriers to entry,
and by extension, power.
These unique attributes of data under surveillance capitalism generate a similarly unique form of
market competition, where the economic value of potentially gaining "total certainty" over Furthermore, corporate hegemony under surveillance capitalism is unique because the very ones
consumer behavior, and in turn, increasing profits drastically makes the accumulation of data a who pioneer the surveillance capitalist logic are also the predominant innovators in many facets
new form of power. It is powerful because access to this vacuum of wealth is controlled strictly of technological research. In an age where information technology is at development’s core, these
by firms that are first adopt this business model and secure high barriers of entry against other firms can hence employ their expertise, invest in R&D using their vast capital, and utilize their
potential competitors. The most prominent example of this is perhaps Google. By being influential market power to integrate the logic of data extraction as an integral component of
essentially the pioneer in mining data for "targeted advertising", it accessed abnormal margins of future industries— for example, Google currently in most markets for “smart products”. 16 Even
profit when the company went public in 2004, where revenues increased by 3,590 percent.10 Yet for large corporations like Apple, who don’t necessarily extract data from customers in such a
this concentration of data and wealth was limited to those who had access to "historical search manner directly 17, it is fair to say that many of its products still require the integration of data
query logs and their corresponding search result clicks",11 which improved the accuracy of search extraction technologies and services pioneered by surveillance capitalists in order to compete in
results up to 31%12 ——a lopsided competitive advantage that Google had possessed (and still its own market. Examples include the integration of popular browser search engines like Google
considerably capitalizes upon) as the pioneer of such technology. Without access to past pools of Chrome (runs on third-party search algorithms) in Apple products, Apple’s development of Apple
behavioral data, even corporate giants such as Microsoft (which owns the search engine Bing) Maps (which runs on first-party data algorithms) and so forth, which all show the expansive
still find it hard to compete with Google simply due to their late entry into the search engine influence of the data extraction logic in traditionally non-surveillance capitalist sectors. This
market—even with acknowledgement of the Microsoft-Yahoo Search Deal, which had already means that surveillance capitalists, by virtue of their implications on competition and broader
allowed Microsoft access to Yahoo's considerable amount of past user search behavior data. As market conditions, are able to guarantee their relevance to information civilization to an
shown, the market for data extraction is governed by high barriers to entry that prevent new firms unprecedented extent. This allows the corporate hegemony under surveillance capitalism to also
from competing against immensely powerful and established surveillance capitalists, even if they be sustained on much longer terms.
have better algorithms.13
Therefore, because a. the market for data extraction as observed currently is so unprecedently
This is notable for consideration, as previous forms of capitalism do not necessarily entail exclusive in nature and b. so potent in its implications on free market economies at large, data
corporate hegemony in the same way that surveillance capitalism ensures its beneficiaries. For and its particular method of extraction has become an economic imperative that is both
industrial14 and managerial15 capitalism, for example, which places power in those who control
3.) unique in how it entails long-term corporate power for beneficiaries, which is hardly possible
9
to undermine (as of currently)
Power.
10
Shoshana Zuboff, “You Are Now Remotely Controlled”. and
11
Kira Radinsky, “Data Monopolists Like Google Are Threatening the Economy”. 2.) intrinsic to surveillance capitalism, where as a result of the previous statement, one cannot
12 compete on a viable surveillance capitalist business model if one is unable to strictly rely upon,
Eugene Agichtein, Eric Brill, and Susan Dumais, “Improving Web Search Ranking by Incorporating
User Behavior Information”.
16
Josef Drexl, “Designing Competitive Markets for Industrial Data - Between Propertisation and Access.”
13
Kira Radinsky, “Data Monopolists Like Google Are Threatening the Economy”.
17
14
Shoshana Zuboff, “You Are Now Remotely Controlled”. Evgeny Morozov, “Capitalism's New Clothes.”
15
Alfred D. Chandler. "The Emergence of Managerial Capitalism."
199 200
duplicate, and perfect the logic of data extraction and hegemonic corporate narrative that 2.) the violation of a consumer’s autonomy is an intrinsic outcome when data is used for
predecessors have pushed for. commercial practices because a consumer can't be said to have genuine control over their own
appetite when data extraction provides firms with the knowledge of how to trigger that appetite
and alter it. Taking Google’s targeted advertising techniques as an example22, a consumer might
3、FREEDOM AS AUTONOMY feel that the product displayed is what they actually want, but that doesn’t change the fact that it
is in fact a product of external imposition made possible by data analytics and undergone by a
Given such a portrayal of surveillance capitalism, it is clear to see that surveillance capitalism’s profit-driven actor. For such consumers, they cannot be said to be truly autonomous.
most quintessential characteristics are underpinned by data extraction. Then, it is important to ask
if there is anything wrong with the act of extraction of data itself. This is quite crucial, since if In response, some may argue that such a violation is contingent on the latter act of data being
there is something wrong with data extraction, then there is naturally something wrong with used in some way23. If data was not used to actually distort a consumer’s appetite, the extraction
surveillance capitalism as well, as its business model cannot be conceived without data. of data is not necessarily problematic24. However, firms will act upon the data they gain by virtue
of surveillance capitalism’s very economic logic. As elaborated upon in the previous section, if
a firm wishes to be competitively viable and profitable as a surveillance capitalist, it must
In response, I argue that the extraction of data is wrong because it simultaneously strips away the
replicate all factors of the business model entailed by its pioneers and predecessors in order to
autonomy of consumers. Here, autonomy18 is interpreted as the capacity to a.) obey oneself 19 and
compete with established competitors. That is to say, if a firm is a surveillance capitalist, it must
b.) control one's own appetite.
use the data it extracts to shape and manipulate consumer appetite; otherwise, the firm is either
rendered uncompetitive in such a market or not a surveillance capitalist at its core.
To clarify, ‘obey oneself’ is a conception of subjective autonomy, where a person is acting upon
their own reasons and motives in order to live a life that does not feel restrained. In this view, if
This is paramount to refuting proponents of surveillance capitalism, who often claim that the
there must be any form of obedience to another authority, that obedience must be self-imposed.
benefits of personalization that data extracting strategies allow could in fact improve the standard
If obedience is self-imposed, that person will not feel bounded or inhibited. For example, if a
of living for customers that are considered favorable for consumption25. As long as firms are
citizen obeys civil laws and norms of a community so that they can enjoy the benefits of
inherently inclined to distort these consumers’ awareness of what is genuine, there is no guarantee
citizenship and legal protection in exchange, the citizen can still be subjectively autonomous in
that the higher ‘standard of living’ that is conceived by consumers doesn’t eventually become a
this case, since their obedience is self-imposed and subjectively empowering20. On the other hand,
mere construct or illusion. To consumers who are looked upon favorably because of their capacity
to ‘control one’s own appetite’ is interpreted as a form of objective autonomy. In this view, for
or tendency to consume, the promotion and display of data’s goodness is implied to run
someone to be autonomous, it is not enough for that person to subjectively feel unrestrained. In
simultaneous to the underlying corporate agenda that aims to nudge further consumption,
order for someone to truly obey themselves, the reasons and motives that one obeys must also be
whereby consumers, ideally, will routinely and indefinitely consume in bulk and rely on data-
authentic and independent, where they shouldn’t be the result of “distortion” by an external
driven services irrespective of whether there is an authentic desire to actually consume in the
source.21 For example, addicts subject to mindless consumption might not necessarily ‘feel’ like
future. This trend leads to an inevitable shift in the dynamics of capitalism from one centered
they are obeying someone else, but their actions may still be objectively a product of an appetite
around fulfilling genuine consumer needs to one in which firms can afford to pursue profit-
or addiction that they don’t have genuine control over. For these addicts, their lives can’t be said
maximization via harnessing consumer appetite in a way that is irrespective of their genuine
to be truly self-driven. Hence, from this two-part view of autonomy, the capacity to a.)
desires.
subjectively obey oneself and b.) objectively control one’s own appetites are both essential to
maintaining a consumer’s freedom as autonomy.
However, why this violation of autonomy should be ethically unacceptable has so far only been
implied rather than addressed explicitly. Why is the quality of being self-driven and autonomous
Understood in this sense, the extraction of data under surveillance capitalism may strip away the
such an ethical imperative?
autonomy of consumers due to the implications data has on consumers’ ability to maintain control
over their needs for consumption.
To respond, maintaining autonomy in itself is necessary to leading and living a good and fulfilling
life. It is necessary because only by having self-driven appetites and genuine needs can one also
Since it was previously established that the extraction of data —in its highly precise and accurate
begin to accordingly define fulfillment for oneself and progress towards that goal; when
form that tells firms how to make a particular consumer want and choose— is intrinsic to
mindlessly consuming, an individual is not being fulfilled. As afore-explained, if firms are able
surveillance capitalism, it follows that
to alter consumer appetite, consumers are dependent on firms in determining what is satisfying
and when one is actually satisfied. And given that firms inherently profit off of never-ending
consumer desire and are invested into sustaining it, it is unconceivable why consumption driven
18
From ‘autonomos’, from autos ‘self’ and nomos ‘law’. Contrasts with heteronomy, ‘hetero-’ meaning by corporate influence via the use of data to alter consumer behavior will result in consumers’
‘other/another/different’ and ‘-nomy’ from nomos. genuine fulfillment as opposed to the generation of more desires, and subsequently, more needs.
19
Joel Feinberg, “The Moral Limits of the Criminal Law Volume 3: Harm to Self”.
20
People may bound themselves to society and its norms “to find a form of association that will defend 22
Shoshana Zuboff, “You Are Now Remotely Controlled”.
and protect the person and goods of each associate with the full common force, and by means of which 23
Titus Stahl, “Indiscriminate Mass Surveillance and the Public Sphere.”
each, uniting with all, nevertheless obey only himself and remain as free as before.” See Jean-Jacques 24
This view is elaborated upon and discussed later on from the lens of Freedom as Non-Domination.
Rousseau and Richard W. Crosby, Of the Social Contract. Pp. 49-50.
25
21
John Christman, “Autonomy in Moral and Political Philosophy”. Frank Pasquale, The Black Box Society.
201 202
As such, consumers who are driven by external and superimposed needs to consume will only be Thus, it might not be the case that violation of consumer autonomy is a unique phenomenon to
further away from fulfillment. Hence, it can be concluded that in this market context, autonomy surveillance capitalism.
of consumers is a prerequisite to ever achieving self-fulfillment; subsequently,
To rebut, one can argue that in previous cases of capitalism, the violation of consumer autonomy
1.) by breaching autonomy, the act of data extraction is ethically unacceptable, was a byproduct of a dysfunctional 30 system rather than its intentions: previous forms of
since it prevents us from living a fulfilling life. capitalism did not conceive the violation of consumers’ autonomy as a necessary component of
a firm’s business model, nor was it essential to the broader economic order; the system may still
have had that objective of serving consumer needs. Instances of abuse are thus not intrinsic, but
rather, at best habitual as a result of perverse applications of marketing strategies. Contrarily, data
While the first two prongs of the criteria (1. ethically unacceptable and 2. intrinsic to surveillance extraction and the violation of consumer autonomy that inevitably follows is engrained into the
capitalism) are met, it isn't immediately apparent why the violation of consumer autonomy is very logic of surveillance capitalism. Granting that to be true,
unique to surveillance capitalism.
3.) perhaps one can claim that the logic of creating and 'controlling' consumer
One may raise the objection that the manipulation of consumer needs irrespective of genuine need appetite is only fully actualized in surveillance capitalism and is thus unique to this
has been a strategy that firms have employed way before the emergence of surveillance capitalism extent.
in its current form. For one, firms have long tried to increase likelihood of consumption through
devious means. Studies have shown that firms have developed cognitive strategies such as However, even then, it isn't quite clear why this is deserving of greater concern other than that it
"option-framing" in selling customizable products with add-ons, where they can influence is intrinsically wrong: to acknowledge that the violation of consumer autonomy was observed
"consumers' decision making regarding the total number of finally chosen product options" 26 by before surveillance capitalism is to concede that this new 'wrong', in effect, may just be at best a
presenting information in a way that is linked to a specific mode of informational-processing so culmination of development from the past.
as to take advantage of consumers' bounded rationality in a limited information environment.27
For example, experimental results show that "consumers choose a higher number of options in Therefore, from a discussion of freedom as autonomy and how data extraction violates it, we can
the delete (versus add) frame," where a consumer is more likely to purchase a product with the conclude that there is indeed something intrinsically wrong with surveillance capitalism that also
full set of add-ons if all of them were presented in the "default" option.28 This shows how the happens to be unique to a certain extent. Thus, we have satisfied the first two prongs of our criteria
tendency and ability to exploit consumer irrationality rather than to increase sales by persuasion and began to tackle the third. However, the fourth prong remains unaddressed—how is this wrong
could have been and likely has been in practice without involving data extraction at all. any worse than what we had before? Greater yet, is the violation of autonomy that could occur as
a result of data extraction really our greatest concern? In broader terms, having established that
Even ignoring these specific instances of behavior influencing and instead observing the broader companies will act upon the data they extract, are we ultimately worried because of all the ways
concept of 'genuine needs', are we really only pursuing genuine needs anyway? Take smart in which data can be used (which results in violation of consumers’ autonomy), or because of
devices or the internet—many would very well conceive of these goods and services as how companies are in a position to do so?
indispensable, yet the reality is that they are all super-imposed by an information civilization and
new modes of social interaction that is to a large extent fueled by corporations that develop these The lenses of freedom as non-interference and non-domination may provide us with answers.
technologies.
This sheds insights upon how firms’ intent to impose ‘fake’ needs stems beyond the economic 4、FREEDOM AS NON-INTERFERENCE
logic of surveillance capitalism. Firms act in such a manner not just because they must in cases
of surveillance capitalism, but also because profit maximization is their private economic In the previous section, we have identified violation of autonomy as a ‘wrong’ that occurs under
objective29. As stakeholders wholly invested into maintaining the existence and relevance of the surveillance capitalism. However, it is perhaps only one of many wrongs that can be identified in
market economy as a central allocative mechanism, the desire to sustain consumer needs and relation to the usage of data. This poses the question of how all those wrongs might cumulatively
societal demand infinitely is a natural byproduct. And if the usurpation of consumers’ autonomy reflect something even worse about the state of corporate surveillance.
is a means to do so, it would seem reasonable that firms will gravitate towards such strategies.
To do so, one might try extending the argument of the previous section: consumer will is not just
affected by distorting one’s appetite, but outright and continuously violated in all the ways that a
26
Dipayan Biswas, “The Effects of Option Framing on Consumer Choices: Making Decisions in Rational consumer can be wronged. Essentially, violation of autonomy and all the other wrongs that can
versus Experiential Processing Modes.” pp. 284-299. arise are significant in whole because they all ultimately interfere with what a consumer wants.
27
Daniel Kahneman, “Maps of Bounded Rationality: Psychology for Behavioral Economics.” Pp. 1449-
1475. If this is true, surveillance capitalism can be said to be ethically problematic by the conception of
freedom as non-interference31. In this view, when there is an external force that is acting directly
28
Dipayan Biswas, “The Effects of Option Framing on Consumer Choices: Making Decisions in Rational against a consumer’s will, the consumer is prevented from actualizing what would have been an
versus Experiential Processing Modes.” pp. 284-299.
29 30
Craig Dunn and Brian K. Burton, “Friedman's ‘The Social Responsibility of Business Is to Increase Its See Introduction on interpretation of ‘dysfunctionality’.
31
Profits’: A Critique for the Classroom.” Philip Pettit, Republicanism: a Theory of Freedom and Government.
203 204
actionable desire. As such, they are both objectively and subjectively constrained. Thus, one technologies to improve national security and streamline queues for public services, for
might argue that the consumer is prevented from freely living a good life according to their own example.39 As such, many governments are gaining an awareness of these firms' operations.
will.
However, in these cases, regulation is ultimately contingent on these practices first being exposed
First, how might surveillance capitalists be able to consistently violate consumer will in its to the public or the regulators. In other words, if Facebook was able to conceal that it was
numerous ways? One might say that surveillance capitalists are able to do so via "hidden" extracting data in the manner it did or does, regulation would have still been rendered rather
commercial processes, which originates with the opaque nature of data extraction. 32 If a useless; if Baidu was unwilling to ally itself to the state, the government would not have known
consumer was to be unwilling to give data and data was forcibly extracted from them, there is no what it now knows. This is a crucial premise, as surveillance capitalists are also the very ones
meaningful way to counter that.33 The lack of recourse stems beyond the extraction of data itself, that pioneer technologies—not necessarily just data extraction technologies, but often tech on
since the opacity implies that even traditional activities that involve the interference with many fronts (as explained in previous sections). In combination with market provisions and legal
consumption (such as price discrimination, 34 “market segmentation”, 35 and “behavioral protections for own-brand data technologies, even if current operations are exposed, it is likely
modification” 36 ) all can be done in a way that is undetectable and unobjectionable, unlike that surveillance capitalists will still be able to develop new methods of opaque data extraction
previously—After all, how can one detect data being used if one is oblivious to data being to ensure that their commercial practices remain undetected in the near future. Even if we suppose
collected in the first place? that these firms are dependent on state funding 40 and that states can try transferring this funding
towards competing research efforts against surveillance capitalists, government actors may still
To validate such an argument, we must first prove that corporate surveillance techniques are lack the corporate expertise, organization, and capital to compete on the same platform.
intrinsically opaque. The opacity of data extraction may be intrinsic for two reasons: For one,
data can only be useful if consumers are largely ignorant of its procedures. If consumers are aware 2.) As such, it makes the hidden quality of a surveillance capitalist's operations a
of the goings-on, they are able to develop mechanisms of defense (for example, by seeking legal likely permanent and intrinsic component of its business model so long as surveillance
recourse). At the very least, those consumers will likely consciously or subconsciously alter their capitalism is the economic order.
behavior in some way that is less natural and useful towards a firm's purposes of individualistic
consumer profiling. If optimum individualistic consumer profiling is intrinsic to surveillance Because the hidden nature of these operations allows surveillance capitalists to go undetected,
capitalism, then so must the opacity of data's extraction. Second, data extraction is intrinsically
opaque because there is no meaningful way for even policymakers to combat such behavior, 3.) extraction of data under surveillance capitalism is also for this very reason
unlike before.37 If there are no means of oversight and ways to combat such behavior, profit- unique compared to other methods that firms can use to violate consumers’ autonomy,
driven firms will not naturally subject themselves to regulation, since firms will likely be as governmental scrutiny has traditionally been the predominant mechanism for
incentivized by the extra profit margins that opaque data extraction can entail. constraining corporate behavior.
One might counter that while that may have been the case at one point, in the near future, one can In order to prove the hidden nature of data extraction to be ethically problematic from the lens of
reasonably expect policymakers to develop the capacity to counter surveillance capitalists. This freedom as non-interference, though, it must also be proven that the act of interfering with
can be seen in the increasing scrutiny of major surveillance capitalists such as Facebook by consumer will is necessary for surveillance capitalism to function; We must prove why for all
regulators such as the EU aimed to pressure Facebook into adopting changes.38 It is also not instances, it is intrinsically the case that consumers must stand opposed to hidden commercial
unconceivable that some firms will ally themselves to the state. Baidu, the prominent search practices. This might not yield as clear a conclusion.
engine in China and major proponent of data-analytics and face recognition technologies, is
known for partnering with the Chinese government in providing the data and the extraction-
The most intuitive argument in support would be to utilize the logic of the previous section, where
hidden practices such as data extraction intrinsically violates one’s autonomy and is ethically
unacceptable. Since data extraction violates a consumer’s autonomy, one might say that it is
irrational to tolerate such procedures.
32
Power.
33 However, what is missed is the fact that consumers don't always choose based upon what is
Sarah Myers West, “Data Capitalism: Redefining the Logics of Surveillance and Privacy.” pp. 20-41.
reasonable and apparently rational. Consumers are often subject to a "double-bind", where they
34 are "caught between desires for privacy and the ability to form meaningful communities with
Akiva A Miller, ‘What Do We Worry About When We Worry About Price Discrimination?: The Law
and Ethics of Using Personal Information for Pricing.’ pp. 41-95. other users online without opting out of these services".41 Facebook famously makes "the use of
its social network conditional on its being allowed to limitlessly amass every kind of data
generated by using third-party websites and merge it with the user's Facebook account. These
35
Karen Yeung, “Five Fears about Mass Predictive Personalization in an Age of Surveillance Capitalism.” third-party sites include firstly services owned by Facebook such as WhatsApp or Instagram, and
pp. 258-269.
36
Zuboff, The Age of Surveillance Capitalism: The Fight for the Future at the New Frontier of Power. 39
Jiangping Zhou, Qihao Wang, and Haitao Liu, “Evaluating Transit-Served Areas with Non-Traditional
37
Data: An Exploratory Study of Shenzhen, China.”
Rinie van Est and Joost Gerritsen, Human Rights in the Robot Age. 43.
40
Mariano-Florentino Cuéllar, and Aziz Z. Huq. “Economies of Surveillance.”
38 41
Paul M. Schwartz. "The EU-U.S. Privacy Collision: A Turn to Institutions and Procedures." Sarah Myers West, “Data Capitalism: Redefining the Logics of Surveillance and Privacy.” pp. 20-41.
205 206
secondly websites and apps of other operators with embedded Facebook API.”42 This enormous To clarify, to possess freedom as non-domination is to not be subject to any arbitrary power that
sphere of influence over what are often the "essentials" to how we as people interact, can prevent one from acting upon their will. A power is arbitrary if there are no mechanisms in
communicate, and live in the contemporary day and age means that consumers are often inclined place to limit the extent to and the conditions under which the power can be exercised. One is
to tolerate a firm's data extraction even if it is against their will or coerced. While one might find unfree if subject to arbitrary power because there is no sense of security.46 A slave under the
this ethically problematic for other reasons, freedom conceived as non-interference doesn't take dominion of a master is insecure because the master is able to tyrannize over the slave at any
into account the conditions under which the consumer choice is made, strictly construed. A choice moment; even if the master is benevolent, the duration is indefinite and up to the discretion of the
is a choice made. master.
In fact, even if one grants that a consumer is acting rationally, it isn't necessarily the case that In the context of surveillance capitalism, firms gain arbitrary power since data extraction places
they will object to data extraction because it violates their autonomy. For example, most websites firms in a position to disregard the consumer. As afore-explained in previous sections, by
would ask the user to consent to their privacy settings and use of browser cookies. For these users, extracting data, firms are able to manipulate consumer appetite. In the context of total domination,
it isn't unconceivable that at least some of them will agree, as there are reasonable and perceivable this is problematic because firms gain the power of taking a consumer’s self-control away
benefits 43 such as personalized webpages that comes with accepting those settings, which some irrespective of whether the consumer actually welcomes this intervention. This ethical concern is
users might desire. Similarly, as long as there are any possible rationales for a consumer to agree contingent not on the later act of triggering and altering a consumer's appetite itself--which can
to the hidden data extraction process and the commercial practices that follow, it can't be said manifest in the form of manipulation, “behavioral modification”, and so forth47 -- but on the very
that firms necessarily interfere with consumer will. Thus, surveillance capitalism may not be ability to do so, which is entailed by data extraction. This is because once a firm possesses data
ethically unacceptable by the standard of freedom as non-interference. and is in a position to alter consumer appetite, the notion that only consumers can alter and choose
what they need is lost. Consequently, firms also lose respect for a consumer’s will. Under such
Of course, one may object that this is similar to subjective autonomy, where the consumer feels conditions, firms with the economic-objective of profit-maximization48 have no reason not to
that they are acting upon their own will even though it could be in reality a guided and externally capitalize upon their newly gained advantage.
influenced response. However, if consumers feel “empowerment” as a result44, why must this be
ethically wrong? If a consumer’s fulfillment in life can be subjectively determined, why can’t This distinction between the "position to" and the "act of" altering consumer appetite is important
one’s subjectively perceived satisfaction be legitimate as well? A consumer may still be able to because while many scholars like Zuboff talk about what is wrong with the following acts or
live a subjectively satisfying life without having all needs fulfilled. In fact, without the ways in which data can be or is used, they don't elaborate upon why the appropriation of data in
conveniences that can be offered data-driven services45, a consumer may perceive themselves to this form is in itself something intrinsically bad49. Hence, critics such as Morosov assert that
be worse off. Then, from a consideration of the various ways in which data can be used, Zuboff might as well rename "surveillance capitalism" to be, for example, “behavior modification
surveillance capitalism might even offer various considerable consumer-benefits---even if its capitalism” since it appears that the "latter is [her] real object of concern".50 This can be addressed
commercial practices may be worryingly hidden from the public eye. by considering data extraction as a feature of surveillance capitalism is ethically wrong due in the
very position to use data (which is inherent in the stage of data appropriation) and not just wrong
by the potential ways that data can be used.
5、FREEDOM AS NON-DOMINATION 2.) Since these corporations’ arbitrary power is driven by the hidden nature of
these commercial practices, which has been proven in the previous section to be intrinsic,
What, then, is so concerning about firms being able to use data extracted from consumers? this position to dominate must also be an intrinsic outcome of surveillance capitalism.
Perhaps, one might be concerned because of the very fact that firms are in a position to use data, Certainly, traditional forms of corporate power over consumers did exist, such as how "Apple
where data extraction is problematic not because of what can follow, but because it is regularly pushes customers around, even preventing from using third-party repair services.”51
characteristic of an arbitrary power that surveillance capitalists gain through the process of data However, because corporate actions were traditionally detectable, that form of power could be
extraction; firms accumulate not just market power, but the capacity to facilitate social control as restrained—it didn't need to be arbitrary to the fullest extent. Essentially, while capitalism in its
well—which is wrong in and of itself from the view of freedom as non-domination. traditional forms may tend to favor non-regulation, regulation was still possible if needed.
Now, however, not only is the consumer insecure against the surveillance capitalist, the consumer
is unaware of their insecurity and government regulators are unable to regulate these opaque
42
commercial procedures or develop methods of regulation because a.) only surveillance capitalists
“Press Release: Preliminary Assessment in Facebook Proceeding: Facebook’s Collection and Use of
control access to the technologies and means of further developing them, b.) it is largely up to
Data from Third-Party Sources is Abusive.”
46
Philip Pettit, Republicanism: a Theory of Freedom and Government.
47
43 Power.
Mariano-Florentino Cuéllar, and Aziz Z. Huq. “Economies of Surveillance.” 48
Craig Dunn and Brian K. Burton, “Friedman's ‘The Social Responsibility of Business Is to Increase Its
Profits’: A Critique for the Classroom.”
44 49
Shoshana Zuboff, “You Are Now Remotely Controlled.” Evgeny Morozov, “Capitalism's New Clothes”.
45 50
Karen Yeung, “Five Fears about Mass Predictive Personalization in an Age of Surveillance Capitalism.” Ibid.
51
pp. 258-269. Ibid.
207 208
firms to expose themselves and c.) there is no meaningfully instituted obligation or leverage by 2. intrinsic to surveillance capitalism
which to compel firms to do so (as elaborated upon in the previous section). This not only and to be worse than previous or existing forms of capitalisms,
reinforces the ability to extract data indefinitely in the marketplace, but implies that the same 3. a unique feature to surveillance capitalism
corporate power can be applied to social spheres and political domains without effective 4. worse in terms of societal impact in general
resistance—all of which done with the right of total discretion vested in the firm.
To focus my analysis, I discussed surveillance capitalism’s ethical status in terms of the
3.) This accumulation of arbitrary power is therefore unique to surveillance philosophical concept of freedom, as understood in three different ways: freedom as autonomy,
capitalism in comparison to previous forms of capitalism. non-interference, and non-domination:
In an exploration of freedom as autonomy, I have proven that the extraction of data under
Moreover, the lack of recourse against surveillance capitalists’ arbitrary power in executing its surveillance capitalism prevents a consumer from being autonomous, where they are unable to
commercial procedures makes our societal norms and values vulnerable to being undermined. truly obey themselves and maintain genuine control over their own appetite. I have shown how
For example, a study using an automated tool that runs browser-based experiments regarding the this usurpation of autonomy necessarily prevents consumers as individuals from pursuing and
relation between user behavior, Google ad settings, and the ads displayed to users found that for living a good and fulfilling life, where this violation is 1.) ethically unacceptable, 2.) intrinsic to
the high-paying job ads subject to experimentation, Google's targeted advertising techniques had surveillance capitalism, and 3.) unique to surveillance capitalism compared to existing forms of
led to the ads being shown "1852 times to the male group but just 318 times to the female capitalism to a limited extent.
group".52 As a regular female user, one would not even be aware of this disparity, even though it In a discussion of freedom as non-interference, I tried to extend the previous section’s discussion
is quite problematic for obvious reasons. Data extraction for targeted-advertising in ways such as of autonomy to a broader exploration of how the vast number of ways in which data can be used
this demonstrates just how social inequalities and injustices can be exacerbated by corporations, are cumulatively and unprecedently harmful and unacceptable. Although this section did not yield
yet will go inevitably unaddressed because of the opacity of surveillance capitalists’ operations. definitive conclusions, it provided insights into to the nuances behind support for surveillance
capitalism, where it could provide considerable consumer benefits in some instances.
In an investigation of freedom as non-domination, I shifted away from analyzing surveillance
Perhaps it may appear that “nary a day goes by when we do not gain in some material way from
capitalism based on the ways in which data can be applied, and instead, evaluated how
the extraction and deployment of” data,53 yet it can’t be neglected that a sense of security in the
surveillance capitalism might be problematic in the act of appropriating data itself. In considering
values and moral decency of our societal environment is also essential for us as moral beings to
the unprecedented arbitrary power that surveillance capitalists are able to accumulate during the
lead a good life.54 This implies that the violation of consumer autonomy which data extraction
appropriation-stage, I have shown how the corporate dominion that underlies corporate
entails (as aforesaid) is not just wrong because the power to do so is arbitrary,
surveillance manifests in such a way that is 1.) Ethically unacceptable, 2.) intrinsic to surveillance
capitalism, 3.) unique, and 4.) with dangerous societal implications.
4.) but also much worse in its effects under surveillance capitalism compared to what we had This allowed me to finally conclude that the same features of surveillance capitalism that had led
faced before. This is because the logic and methods behind surveillance capitalism does not to the usurpation of consumers’ freedom as autonomy manifests in a 4.) far worse manner as
merely actualize a perverse corporate desire for profit-maximization that has been culminating opposed to what occurs under other forms of capitalism due to the emergence of corporate
for all this time, but its business model also increasingly equips corporations with the ability to domination it makes possible. Even if the usurpation of autonomy is not to be held against
disregard the moral grounds that constitute much of modern liberal society, and ultimately, surveillance capitalism, there will still be another identified feature in this paper that makes this
human nature. new economic order wrong and far worse.
Therefore, even if one is not to buy the violation of consumer autonomy as an intrinsic wrong in That being said, there are various limitations to this paper. Namely, that all the normative
the economic order, there must still be something wrong and much worse in surveillance judgments made are premised upon “surveillance capitalism” indeed being exactly the same as
capitalism because of its implications on the creation of arbitrary social and economic power, described in this paper. However, although the depiction of surveillance capitalism in this paper
where the emergence of surveillance capitalism actualizes a tyrannical corporatocracy that has may be true, it is conceded that it may only represent a narrow conception of what surveillance
no meaningful obligation to the advancement and preservation of our humanity. capitalism can be; given that “surveillance capitalism” is still a new and novel notion at the time
of writing, we may yet to have understood what we are truly confronted with. “Surveillance
capitalism” can potentially refer to much more to what is underpinned by this paper, such as in
the gradual integration of portions of the surveillance capitalist business model into the healthcare
6、CONCLUSION sector and so forth, where data may be used in much more distinct ways due to the differing
natures and structures of these markets 55 . This may entail more subtle motives and precise
In this paper, I assessed whether there is something wrong with surveillance capitalism in terms regulatory frameworks in these specific cases that are not fully considered in this paper. The same
of a four-pronged metric, where the identified wrong must be: concerns that are brought up in this paper, then, may not apply as the concept of “surveillance
1. ethically unacceptable capitalism” develops to include much more territory in the future.
52
Nonetheless, this paper does have several useful implications, which can be developed in future
Amit Datta, Michael Carl Tschantz, and Anupam Datta, “Automated Experiments on Ad Privacy
studies and research. First, such an ethical critique in this paper alludes to a potential exploration
Settings.” pp. 92-112.
of the social responsibilities of corporations as well as how those responsibilities might differ for
53
54
Francisco J. Ayala, “The Difference of Being Human: Morality.” pp. 9015-9022.
55
209 210
firms in this day and age given the changing circumstances of an information civilization. Second,
this paper’s concerns pertaining to the role of state scrutiny in restraining corporate action invites Economics,” American Economic Review 93, no. 5 (2003): pp. 1449-1475,
further inquiries regarding the dynamics of corporate-state relationships, which would constitute
a much more detailed normative investigation of how our political and economic domains should https://doi.org/10.1257/000282803322655392).
interact.
Dipayan Biswas, “The Effects of Option Framing on Consumer Choices: Making Decisions in
Despite the critical approach adopted in this paper, it should also be noted that this paper is by no
means pessimistic; although surveillance capitalism has been identified in this paper to be Rational versus Experiential Processing Modes,” Journal of Consumer Behaviour 8, no. 5
problematic for various reasons, it should not be taken that society is beyond repair. Rather, it is
merely implied that now that we have more insights into grasping what surveillance capitalism (2009): pp. 284-299, https://doi.org/10.1002/cb.288).
might be, although we may still be oblivious to much of what is going on, we ought not to be
willfully ignorant to the trends that we do see. It is by gaining this awareness as consumers,
Eugene Agichtein, Eric Brill, and Susan Dumais, “Improving Web Search Ranking by
individuals, and governments that we can start making the first small steps towards bringing
change in the economic domain for the better.
Incorporating User Behavior Information,” ACM SIGIR Forum 52, no. 1 (2019): pp. 11-
18, https://doi.org/10.1145/3308774.3308778).
Evgeny Morozov, “Capitalism's New Clothes,” The Baffler, February 6, 2019,
https://thebaffler.com/latest/capitalisms-new-clothes-morozov).
Bibliography
Francisco J. Ayala, “The Difference of Being Human: Morality,” Proceedings of the National
Akiva A Miller, ‘What Do We Worry About When We Worry About Price Discrimination?:
Academy of Sciences 107, no. Supplement_2 (May 2010): pp. 9015-9022,
The
https://doi.org/10.1073/pnas.0914616107).
Law and Ethics of Using Personal Information for Pricing’ (2014) 19 Journal of
Frank Pasquale, The Black Box Society (Harvard University Press 2015).
Technology Law & Policy: 41-95.).
Jean-Jacques Rousseau and Richard W. Crosby, Of the Social Contract (Brunswick, OH: Kings
Alfred D. Chandler. "The Emergence of Managerial Capitalism." The Business History Review
Court Communications, 1978)). Pp. 49-50.
58, no. 4 (1984): 473-503.
Jiangping Zhou, Qihao Wang, and Haitao Liu, “Evaluating Transit-Served Areas with Non-
Amit Datta, Michael Carl Tschantz, and Anupam Datta, “Automated Experiments on Ad
Privacy
Traditional Data: An Exploratory Study of Shenzhen, China,” Journal of Transport and
Settings,” Proceedings on Privacy Enhancing Technologies 2015, no. 1 (January 2015):

Land Use 11, no. 1 (2018), https://doi.org/10.5198/jtlu.2018.1296).
pp. 92-112, https://doi.org/10.1515/popets-2015-0007).

Joel Feinberg, “The Moral Limits of the Criminal Law Volume 3: Harm to Self” 3 (1989),
Craig Dunn and Brian K. Burton, “Friedman's ‘The Social Responsibility of Business Is to https://doi.org/10.1093/0195059239.001.0001).
Increase Its Profits’: A Critique for the Classroom,” January 2006, John Christman, “Autonomy in Moral and Political Philosophy,” Stanford Encyclopedia of
https://www.researchgate.net/publication/288713811_Friedman's_The_Social_Respon Philosophy (Stanford University, January 9, 2015),

sib
https://plato.stanford.edu/entries/autonomy-moral/#BasDis).
ility_of_Business_is_to_Increase_Its_Profits_A_Critique_for_the_Classroom)
Josef Drexl, “Designing Competitive Markets for Industrial Data - Between Propertisation and
Daniel Kahneman, “Maps of Bounded Rationality: Psychology for Behavioral
211 212
Access,” SSRN Electronic Journal, 2017, https://doi.org/10.2139/ssrn.2862975). Sarah Myers West, “Data Capitalism: Redefining the Logics of Surveillance and
Karen Yeung, “Five Fears about Mass Predictive Personalization in an Age of Surveillance Privacy,” Business & Society 58, no. 1 (May 2017): pp. 20-41,
Capitalism,” International Data Privacy Law 8, no. 3 (January 2018): pp. 258-269, https://doi.org/10.1177/0007650317718185).
https://doi.org/10.1093/idpl/ipy020).
Shoshana Zuboff, The Age of Surveillance Capitalism: The Fight for the Future at the New
Kira Radinsky, “Data Monopolists Like Google Are Threatening the Economy,” Harvard
Frontier of Power (New York: Profile Books, 2019)).
Business Review, March 31, 2015, https://hbr.org/2015/03/data-monopolists-like-google-
Shoshana Zuboff, “You Are Now Remotely Controlled,” The New York Times (The New York
are-threatening-the-economy).
Times, January 24, 2020),
Mariano-Florentino Cuéllar, and Aziz Z. Huq. “Economies of Surveillance,” (2020). https://www.nytimes.com/2020/01/24/opinion/sunday/surveillance-capitalism.html).

https://harvardlawreview.org/2020/02/economies-of-surveillance/.
Titus Stahl, “Indiscriminate Mass Surveillance and the Public Sphere,” Ethics and Information
Paul M. Schwartz. "The EU-U.S. Privacy Collision: A Turn to Institutions and
Technology 18, no. 1 (March 16, 2016): pp. 33-39, https://doi.org/10.1007/s10676-016-
Procedures." Harvard Law Review126, no. 7 (2013): 1966-2009. Accessed March 10,
2020. www.jstor.org/stable/23415063). 9392-2).
Philip Pettit, Republicanism: a Theory of Freedom and Government (Oxford: Oxford

University
Press, 2010)).
“Press Release: Preliminary Assessment in Facebook Proceeding: Facebook’s Collection and
Use of Data from Third-Party Sources is Abusive.” Bundeskartellamt. Accessed March
10, 2020.
https://www.bundeskartellamt.de/SharedDocs/Publikation/EN/Pressemitteilungen/2017/1
9_12_2017_Facebook.pdf?__blob=publicationFile&v=3).
Rahel Jaeggi, “What (If Anything) Is Wrong with Capitalism? Dysfunctionality, Exploitation
and Alienation: Three Approaches to the Critique of Capitalism,” The Southern Journal
of Philosophy 54 (2016): pp. 44-65, https://doi.org/10.1111/sjp.12188).
Richard Kraut, “Aristotles Ethics,” Stanford Encyclopedia of Philosophy (Stanford University,
June 15, 2018), https://plato.stanford.edu/entries/aristotle-ethics/).
Rinie van Est and Joost Gerritsen, Human Rights in the Robot Age (Rathenau Institut, 2017)
43.
213 214

Horizon Academic Research Journal Vol. 1 First Edition - Book View

Uploaded by

Copyright:

Horizon Academic Research Journal Vol. 1 First Edition - Book View

Uploaded by

Document Information

Copyright

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Horizon Academic Research Journal Vol. 1 First Edition - Book View

Uploaded by

Copyright:

The Horizon Academic

The Effects of the Coronavirus Pandemic on Student Learning Anika Khurana 01

Melanoma Diagnosis using Convolutional Neural Networks Ansh Chaurasia 14

Copyright 2021, Sunrise International Education Inc.

Oncogenic Viruses: A Potential CRISPR Treatment Neha Matai 50

A Project of Sunrise International Education Inc.

Anxiety and Procrastination: What is the Association? Hriday Nayak 109

electronic, mechanical, photocopying, recording, Function of Aerobic Exercise

What, if Anything, is Wrong with Surveillance Capitalism? Aaron Chen 196

The connectedness of today’s world enables us to unlock the boundless

5 Underprivileged Learners 6 Discussion

Figure 2: Diagram of the ResNet50 Architecture

In essence, the ResNet50 architecture functions by stacking up sets of the

Figure 3: Diagram of EfficientNet Architecture Used

[noaa] American cancer society. https://cancerstatisticscenter.cancer.org/#!/cancer-

[noac] Dermoscopy. https://dermnetnz.org/topics/dermoscopy/. Ac-

[noad] Melanoma statistics. https://www.curemelanoma.org/about-

[noae] SIIM-ISIC melanoma classification.

“Status Quo Bias.” BehavioralEconomics.com — The BE Hub, 28 Mar.

Figure 6: Extracurricular Advice Websites

Figure 8: The time of task insurance

3.3 A Potential CRISPR Treatment for HPV

Figure 3: Playoff Prediction Accuracy of Self-made Prediction Model

2.2 Exposition of the Machine Learning Models

Figure 5: Playoff Prediction Accuracy of Different Machine Learning Models

2.2.3 Home and Away Investigation

Figure 6: 2018 LR Prediction Results

def ensive rebounds ∗ 100

yA − y B 4.3 Machine learning prediction models

[Aut09] No Author. Machine learning - logistic regression.

[Dhi03] K Dhiraj. Top 5 advantages and disadvantages of decision

[Nar04] Kumar Naresh. Advantages and disadvan-

[TA11] Ibrahem Abdlehameed Hassanien Aboul Ella Tharwatt Alaa, Abstract

5.4 Racial Bias

[TS89] B. Tuckman and T. Sexton. Effects of relative feedback in over-

[vE03] W. van Eerde. A meta-analytically derived nomological network

[VFGI12] S. Vahedi, F. Farrokhi, F. Gahramani, and A. Issazadegan.

5.1.2 Prioritizing the most pathogenic variants with 2 different strate-

[FMN] Critical issues in bdnf val66met genetic studies of neuropsychiatric

[JB15] Francielle Mina Michelle Lima Garcez Alexandra Ioppi Zugno

Figure 3: PolyPhen2 score → pathogenicity → Grantham score. For this strat-

Figure 1: Missense mutations of BDNF and BDNF-AS reported in the

Figure 4: NGF domain of two BDNF isoforms. NP 001137282.1 is isoform e of

Figure 7: Sopma results of missense mutation V66M with sequence length

[n.anda] n.a. About folding@home. Folding@home, n.d.

[n.andb] n.a. Designing proteins from scratch. Amgen Science, n.d.

3.3 Aerobic Exercise Increases BDNF

Entrainment can be described as the locking of frequencies between two sepa-

[C. 13] C. Bosch-Bouju, B. I. Hyland, and L. C. Parr-Brownlie. Motor tha-

Karniel, Yuval, Amit Lavie-Dinur, and Tal Azran. "Broadcast coverage of

Evgeny Morozov, “Capitalism's New Clothes,” The Baffler, February 6, 2019,

Settings,” Proceedings on Privacy Enhancing Technologies 2015, no. 1 (January 2015):

pp. 92-112, https://doi.org/10.1515/popets-2015-0007).

https://www.researchgate.net/publication/288713811_Friedman's_The_Social_Respon Philosophy (Stanford University, January 9, 2015),

Mariano-Florentino Cuéllar, and Aziz Z. Huq. “Economies of Surveillance,” (2020). https://www.nytimes.com/2020/01/24/opinion/sunday/surveillance-capitalism.html).

2020. www.jstor.org/stable/23415063). 9392-2).

Philip Pettit, Republicanism: a Theory of Freedom and Government (Oxford: Oxford

“Press Release: Preliminary Assessment in Facebook Proceeding: Facebook’s Collection and

Use of Data from Third-Party Sources is Abusive.” Bundeskartellamt. Accessed March

of Philosophy 54 (2016): pp. 44-65, https://doi.org/10.1111/sjp.12188).

Richard Kraut, “Aristotles Ethics,” Stanford Encyclopedia of Philosophy (Stanford University,

June 15, 2018), https://plato.stanford.edu/entries/aristotle-ethics/).

You might also like