Comparison of Parallelized Sequence

Alignment Algorithms
CSE4001 – Parallel and Distributed Computing

PROJECT BASED COMPONENT REPORT

By

SHASHWAT NEGI
(17BCE0322)
YASH BHOJWANI
(17BCE0614)
RACHANA REDDY
(17BCE0278)

School of Computer Science and Engineering

March 2019

Page 1
 DECLARATION

I hereby declare that the report entitled “Comparison of Parallelized

Sequence Alignment Algorithms” submitted by me, for the CSE4001 Parallel and
Distributed Computing (EPJ) to VIT is a record of bonafide work carried out by me
under the supervision of Dr.Narayanan Prasanth.
I further declare that the work reported in this report has not been submitted and will
not be submitted, either in part or in full, for any other courses in this institute or any
other institute or university.

Place : Vellore
Date : 5/11/19

Signature of the Candidate

Shashwat Negi

Signature of the Candidate

Yash Bhojwani

Signature of the Candidate

Rachana Reddy

Page 2
 CONTENTS Page
No.

Acknowledgement 2

Abstract 7

Table of Contents 3

List of Figures 4

List of Tables 5

Abbreviations 6

1 INTRODUCTION 8
1.1 Objective 8
1.2 Motivation 8

2 LITERATURE SURVEY 9

3 TECHNICAL SPECIFICATION 12

4 DESIGN (as applicable) 15

5 PROPOSED SYSTEM 17

6 RESULTS AND DISCUSSION 19

7 CONCLUSION 20

8 REFERENCES 21

APPENDIX .

Page 3
 List of Figures

Figure Title Page

No. SNo.
1 Track back of 12
Alignment
2 Parallel Implementation 15
of Smith-Watermann
Algorithm
3 Parallel Implementation 15
of Needleman-Wunsch
Algorthim
4a) Executing SW 16
4b) Executing SW
5 Executing NWomp.c 17
6 Graph 18

Page 4
 List of Tables

Table No. Title Page No.

1 Execution Time Comparison 18

Page 5
 List of Abbreviations

S-W Smith-Watermann
N-W Needleman-Wunsch
OMP Open Multi-Processing
FPGA Field Programmable Gate Array
MPI Message Passing Interface
DNA Deoxyribo Nucleic Acid

Page 6
 ABSTRACT

Parallel and Distributed computing is the future of technology. All products and their
fundamental concepts are being shifted to a parallel computing model. Everybody would
agree that serial computing is easy to implement and use, but simply not efficient enough for
industry level purposes. Due to this reason, day by day higher number of industries are
providing and using cloud solutions which work on the basis on parallel and distributed
computing. For instance, Amazon’s AWS or Google’s
Google Cloud platform are becoming the center for development, may it be in the field of
web development, or in the field of data analytics. To cope up with the fast-paced
improvement in technology, one must also become familiarized with this domain, and hence
this project. Gene sequencing problem is one of the major issues for researchers regarding
optimized system models that could help optimum processing and efficiency without
introduction overheads in terms of memory and time. Bioinformatics and computational
biology is a latest multidisciplinary field which explains many aspects of the fields of
computer science, while computational biology harnesses computational approach and
technologies to respond biological questions conveniently.
The libraries used for the same are: <stdio.h>,<stdlib.h>,<math.h>,<omp.h>,<time.h>.
We would be learning mainly how to detect sequences in proteins and nucleic acids and how
to detect them using parallel computing using multiple threads.
We use this inspiration of our project to create something on a smaller scale, but with a large
scope.

Page 7
 1. INTRODUCTION

1.1 OBJECTIVE

To learn and use OpenMP and its related packages:

To learn and use Parallel processing and compare its efficiency with serial execution by
implementing the above mentioned algorithm.
This would be done in parallel and in serial and a comparison between them would be done.
To work in a team and understand the advantages of teamwork by assigning roles for each
member and deadlines for the project, which would help us get suited to a work-based
environment and benefit us in the future.
To improve our report writing and presentation skills through the frequent Reviews conducted
to check on our progress.

1.2 MOTIVATION

As time passes, the world is becoming more and more oriented towards Parallel Computing.
Many of the tasks that were once carried out sequentially are now being carried out in parallel
so as to use resources more efficiently and get faster results. Genome is an emerging field,
constantly presenting many new challenges to researchers in both biological and
computational aspect of application. Research is being done in the Biology with the
application Smith–Waterman algorithm it is possible to process and understand nucleic acid/
protein sequences. Sequence comparison is a very essential and important operation. They
detect similar or identical parts between two sequences called the query sequence and the
reference sequence. The global and local alignments are the most prevalent kinds of sequence
alignment. In global alignment, we find the superior counterpart between parts of the
sequences. On the other hand, local alignment algorithms try to match parts of sequences and
not the entirety of them. Local alignment is faster than global alignment, due to the lack of
need to align the entire sequences. In our project, we would be implementing the Smith-
Waterman (Local sequence alignment) Algorithm for randomly generated nucleotide
sequences in a serial and parallel manner for comparison and analysis. As common sense
suggests, the parallel implementation should execute and provide the same result as the serial
implementation but in a lesser amount of time.

Page 8
 2. LITERATURE SURVEY

High Performance Streaming Smith-Waterman Implementation with Implicit

Synchronization on Intel FPGA using OpenCL

In this paper, Ernst and Zaid have presented OpenCL-based FPGA Smith-Waterman implementation
that employs two key techniques to greatly improve the utilization of its underlying systolic array
architecture. By eliminating centralized control and through the use of Query Buffers, an arbitrary
number of alignments can be in flight at the same time, resulting in utilization close to theoretical
maximum performance. In parallel computer architectures, a systolic array is a homogeneous network
of tightly coupled data processing units called cells or nodes. Field-Programmable Gate Arrays (or
FPGAs), with their flexible and reprogrammable substrate, are a natural fit for a computationally
intensive algorithm such as the Smith-Waterman algorithm. The Query Buffer contains for each
Processing Element a separate queue with query symbols for the alignments it needs to process.
Whenever the Processing Element encounters the new read token, it checks against the query length
to verify if it is active during this alignment; if so, it reads the next query symbol from its queue. Only
the Input Parser and Output Parser communicate with the on-board DDR memory.

Development of DNA Sequencing Accelerator Based on Smith Waterman Algorithm with

Heuristic Divide and Conquer Technique for FPGA Implementation

In this paper, a new approach has been introduced to reduce the complexity of the Smith Waterman
algorithm for FPGA implementation. The technique for the fastest comparison of the two DNA
sequencing using Verilog on the Xilinx ISE 7.1. In this paper, it was proved that Smith Waterman
algorithm based on divide and conquers technique gives better performance than existing
technique.The Smith Waterman algorithm is the complex and heuristic algorithm for the DNA
sequencing. Divide and Conquer technique helps to reduce the complexity of the main structure. In
previous works, the main structure is divided or break-up into few modules called sub-functions and
at cluster based. However, the result is less sensitive. In this technique, the functions involve in Smith
Waterman algorithm were break up into few sub- functions. These parts are prediction of the sequence
score, previous result score and optimization of the sequencing.

A Parallel Implementation of the Smith-Waterman Algorithm for Massive Sequences

Searching

In this paper, Hsien-Yu Liao, Meng-Lai Yin, Yi Cheng has presented a parallel implementation
methodology of the Smith-Waterman algorithm. The power of parallelization lays on the massive
comparisons. When an unknown sequence is compared with different existing sequences, each
comparison is independent and can be performed independently. This observation highlights the
potential of massive parallelism existing in this particular application. Using this method, it was
proven that high efficiency can be achieved.

Page 9
Parallel Processing Cell Score Design of Linear Gap Penalty Smith-Waterman Algorithm

In this paper, the optimize computational processing element for linear smith-waterman algorithm cell
score based on the parallel computational approach is introduced. Two bits optimized comparator
block were used to compare the DNA sequence characters, while the computation block was used to
complete computation towards reducing the numbers of components involved. The complete
architecture is designed and developed in Altera Quartus II version 13.0 and targeted to Altera Cyclone
IV EP4CE115 Field Programmable Gate Array (FPGA). In realizing the important of the logic
functionality of the design, the modular design approach has been adapted to this design. The
simulation over the module and complete architecture were used for validating the design against the
expected result. Finally, the result obtained from the study indicates that the optimized computational
processing element design for linear smith-waterman algorithm cell score based on the parallel
computational approach is feasible to be implemented in the DNA sequence alignment as the
processing element.

Performance Improvement of the Parallel Smith Waterman Algorithm Implementation Using

Hybrid MPI – Openmp Model

In this paper, a hybrid parallel model was introduced that combines both shared and distributed
memory architectures to improve the performance of the Smith waterman algorithm (SW). The hybrid
model uses both MPI and OpenMp as programming techniques for different memory architectures.
Our improved implementation executes a parallel version of SW algorithm with a row wise
computation of the alignment matrix, which mainly optimizes the memory usage. Using different
parallel programming models and their corresponding implementations.

A Comparative Analysis of Smith-Waterman Based Partial Alignment

In this paper, Aruk, Ustek, and Kursun have given a comparative simulation of three Software based
methods in terms of their runtimes and errors in estimated position of the start of the deletion in the
query sequences. As a result, they have found that Binary Partial Align has the lowest error and very
high speed. Partial alignment aims at finding the best splitting of a query sequence into the former and
latter parts such that their matching scores with (different parts of) the reference sequence are mutually
maximized. In this paper, we give a more detailed comparison of the classical SW,
IncrementalPartialAlign, and BinaryPartialAlign. BinaryPartialAlign has better EDPE than the other
two and is faster than IncrementalPartialAlign, making it the best option among the three.

A Novel Structure of the Smith-Waterman Algorithm for Efficient Sequence Alignment

In this paper, a new architecture for SW algorithm has been presented for the implementation of the
matrix fill-up stage in the SW algorithm. The matrix fill-up stage is one of the most time-consuming
operations in the SW algorithm. The conventional implementation of this phase takes four cycles for
computation of each element in the matrix. The newly proposed design reduces this latency to three
cycles. However, the proposed design takes one overhead cycle at the start of the processing for
ensuring the 100% accuracy. With the newly proposed architecture, the SW algorithm achieves up to
25% speedup.

Page 10
An Efficient and High Performance Linear Recursive Variable Expansion Implementation of

the Smith-Waterman Algorithm

In this paper, an efficient and high performance linear recursive variable expansion (RVE)
implementation of the Smith-Waterman (S-W) algorithm and compare it with a traditional linear
systolic array implementation. An efficient and high performance implementation of the S-W
algorithm based on the linear RVE approach and compared it with a traditional linear systolic array
implementation. The linear RVE implementation is efficient in terms of hardware utilization (both
slices and IOBs) and high performance in terms of time consumption (latency). The results
demonstrate that the linear RVE implementation is upto 2.33 times faster than a traditional linear
systolic array implementation at the cost of utilizing 2 times more resources. The results demonstrate
that the linear RVE implementation performs up to 2.33 times better than the traditional linear systolic
array implementation, at the cost of utilizing 2 times more resources.

Page 11
 3. TECHNICAL SPECIFICATION
Package Requirements:

 <stdio.h> The C programming language provides many standard library functions for file
input and output. These functions make up the bulk of the C standard library
 <stdlib.h> stdlib.h is the header of the general purpose standard library of C programming
language which includes functions involving memory allocation, process control, conversions
and others. It is compatible with C++ and is known as cstdlib in C++. The name "stdlib" stands
for "standard library"
 <math.h> The math.h header defines various mathematical functions and one macro. All the
functions available in this library take double as an argument and return double as the result.
 <omp.h> It is a library that allows memory multiprocessing programming in C
 <time.h> In C programming language time.h (used as ctime in C++) is a header file defined
in the C Standard Library that contains time and date function declarations to provide
standardized access to time/date manipulation and formatting.

Major Algorithm Explanation:

Smith-Waterman Algorithm

Smith-Waterman algorithm calculates the local alignment of two sequences. It guarantees to find out
the best possible local alignment taking into account the specified scoring system. This includes a
substitution matrix and a gap-scoring method. Scores consider match, mismatch and substitution. To
measure the comparison between two sequences, a score be calculated as follows:

Given an alignment between sequences S0 and S1, the following values must be assigned, for each
column:

Procedure:

 ma = (+5)
 mi = (-3)
 G = (-4)

[Match] [Mismatch] [Gap]

 Initialization of the matrix and consider two sequences A and B.

 Matrix filling with the suitable scores. The two sequences are set in a matrix form by means
of A+1 columns and B+1 rows. The value in the first row and first column are set to zero.
 The second and essential step of the algorithm is filling to entire matrix. To fill each and every
cell it is important to know the diagonal values.
 Trace back the sequence for an appropriate alignment is trace backing; before that the
maximum score obtained in the entire matrix has to be detected for the local alignment of the
sequences.
It is likely to those maximum scores can be present in one or more than one cell, in such case
there may be option of two or more alignments, and the best alignment can be obtained by
scoring it.
 Tracing back begins from the position which has the highest value, pointing back with the
pointers, consequently find out the possible predecessor, then go to next predecessor and
continue until it reaches the score 0.

Page 12
Example:

Figure 1 Track back of Alignment

Needleman-Wunsch Algorithm

In order to prove that the proposed algorithm is better, we compared this algorithm with
Needleman Wunsch algorithm. The Needleman–Wunsch algorithm is an algorithm used
in bioinformatics to align protein or nucleotide sequences. It was one of the first applications
of dynamic programming to compare biological sequences. The algorithm was developed by
Saul B. Needleman and Christian D. Wunsch and published in 1970. [1] The algorithm
essentially divides a large problem into a series of smaller problems, and it uses the solutions
to the smaller problems to find an optimal solution to the larger problem. [2] It is also
sometimes referred to as the optimal matching algorithm and the global alignment technique.

This method aligns the pair of sequences from end to end. The entire length of the sequence
is taken into account. An optimal score is calculated from the matrix formed using the
maximum similarity of each character using match, mismatch and gap penalty values of the
sequences. The optimal alignment is achieved by trace back of the matrix.

If x and y are strings, where length(x) = n and length(y) = m, the Needleman-Wunsch

algorithm finds an optimal alignment in O(nm) time, using O(nm) space. Smith Watermanez
algorithm is a clever modification of the Needleman-Wunsch Algorithm which still takes
O(nm) time, but needs only O(min{n,m}) space and is much faster in practice.[2] One

Page 13
application of the algorithm is finding sequence alignments of DNA or protein sequences. It
is also a space-efficient way to calculate the longest common subsequence between two sets
of data such as with the common diff tool.

ExistingTool:
There are few existing tools which have a parallel implementation of the Smith-Waterman
algorithm, but the most prominent one is Crustal W [EMBOSS WATER]. Link:
https://www.ebi.ac.uk/Tools/psa/emboss_water/
EMBOSS Water uses the Smith-Waterman algorithm (modified for speed enhancements) to
calculate the local alignment of two sequences.
We can perform the alignment for protein, DNA or RNA sequences.

Page 14
 4. DESIGN

The program flow is given by the following program.

Page 15
 5. PROPOSED SYSTEM

The problem at hand was tackled with a modular approach. Eight functions were constructed,
each of which would be explained as follows:

 nElement – This function is used to calculate the number of elements that have been
found by the Smith Waterman Algorithm. Three conditions are given: One of which
is to find out if the number of elements in the diagonal are increasing, decreasing or
stable.
 calcFirstDiagElement – This function is used to calculate the position of the
maximum scored value in the matrix. This value needs to be found because the
algorithm suggests that the backtracking to find the path should be started from this
particular point.

 similarityScore – This function is used to find out the optimal order of execution
based on three conditions, which are used to calculate the new values of left, upper
and the diagonal elements.

If the diagonal element > maximum element, Move diagonally upwards

If upper element > maximum element, Move upwards

If left element > maximum element, Move leftwards

Every iteration, the values of maximum element is updated and inserting into the
similarity and predecessor matrices.

 matchMismatchScore – This function is used to calculate a similarity function or the

alphabet for a match or mismatch.
If the value of the two elements are equal, it is a match, otherwise it’s a mismatch.
 Backtrack – The purpose of this function is to modify the matrix that needs to be
printed and helps us identify the path that needs to be taken to get the most optimum
solution.
 printMatrix – It’s a looped iteration implementation to display the matrix.
 printPredecessorMatrix - It is in this function in which we print the arrows

depicting the path of local alignment.

Generate – This function generates the two sequences A and B which would be
locally aligned with each other. A random seed is used to ensure the reproducible
nature of the output.

If x and y are strings, where length(x) = n and length(y) = m, the Needleman-Wunsch

algorithm finds an optimal alignment in O(nm) time, using O(nm) space. Smith-waterman's
algorithm is a clever modification of the Needleman-Wunsch Algorithm which still takes
O(nm) time, but needs only O(min{n,m}) space and is much faster in practice.[2] One
application of the algorithm is finding sequence alignments of DNA or protein sequences. It
is also a space-efficient way to calculate the longest common subsequence between two sets
of data such as with the common diff tool.

Page 16
 6. Working
Both the programs have been parallelized using section and for directives only. The number
of threads have been set as same.

Executing SWomp.c (Smith-waterrman Algorithm)

Figure 4a). Executing SW

Figure 4b)Executing SW

The Predecessor Matrix helps in identifying the common sequence

Sequence 1: GTTGCATACCCATTT
Sequence 2: GGCGAGACTACTACA
The common sequence visible is “ GC-A-AC”

Page 17
Executing NWomp.c (Needleman-Wunsch Algorithm )

Figure 5. Executing NWomp.c

We can observe the Similar Sequence with the highest score

GTACGCAAACGGGT--G
G-GCG-AGACTACTACG

Page 18
 7. RESULTS AND DISCUSSION

Figure 6 Graph

By considering sequences of lengths 20,30,60,80,100,120 we found the Execution time of

the Parallel Implementation of Smith-Waterman and Needleman-Wunsch Algorithm

Table1 Execution Time Comparison

Sequence Length Smith-Waterman Time Needleman-Wunsch Time
20 0.0010 0.0010
30 0.0020 0.0020
60 0.0020 0.0030
80 0.0030 0.0050
100 0.0040 0.0110
120 0.0060 0.0130

By applying similar parallel techniques ie using “For”-“Task”-“Section” on both algorithm

we get the above execution time.
The execution time so calculated is the total time to find the optimal alignment between 2
sequences without printing it.
We can conclude from the graph and the table that it is much more beneficial to implement
Smith-Waterman Algorithm in parallel and Needleman-Wunsch due to reduced execution
time.

Page 19
 8. CONCLUSION

As we can see in the above graph, for small lengths of the sequence, Smith-Waterman
and Needleman- Wunsch algorithms tend to give the output in almost the same amount
of time. However, as the length or the number of characters increases, the execution
time for Needleman Wunsch implementation also increases exponentially, hence
forming a steep graph.

The serial implementation of Smith-Waterman Algorithm is consider better in terms

of performance and optimal result as it requires lot less space, considers only positive
value. But real life DNA & gene sequences are very long character sequences of the
order 100s .This creates a necessary requirement for executing the sequencing
algorithm in parallel.

However, the Smith-Waterman implementation remains stable with not a high rise in
the execution time because of the parallel execution of the task with two threads,
making the process faster than other algorithms. Hence, the proposed system will be
useful when dealing with large amount of data and can be used for practical purposes..

Page 20
 8. REFRENECES

1. High Performance Streaming Smith-Waterman Implementation with Implicit

Synchronization on Intel FPGA using OpenCL

2. Development of DNA Sequencing Accelerator Based on Smith Waterman Algorithm with

Heuristic Divide and Conquer Technique for FPGA Implementation

3 A Parallel Implementation of the Smith-Waterman Algorithm for Massive Sequences

Searching

4. Parallel Processing Cell Score Design of Linear Gap Penalty Smith-Waterman Algorithm

5. Performance Improvement of the Parallel Smith Waterman Algorithm Implementation

Using Hybrid MPI – Openmp Model
6. An Efficient and High Performance Linear Recursive Variable Expansion Implementation
of the Smith-Waterman Algorithm
7. A Novel Structure of the Smith-Waterman Algorithm for Efficient Sequence Alignment
8. Performance Improvement of the Parallel Smith Waterman Algorithm Implementation
Using Hybrid MPI – Openmp Model

Page 21