Bioinformatics Toolbox

Users Guide

R2014a

How to Contact MathWorks

Web
Newsgroup
www.mathworks.com/contact_TS.html Technical Support
www.mathworks.com

comp.soft-sys.matlab

suggest@mathworks.com
bugs@mathworks.com
doc@mathworks.com
service@mathworks.com
info@mathworks.com

Product enhancement suggestions

Bug reports
Documentation error reports
Order status, license renewals, passcodes
Sales, pricing, and general information

508-647-7000 (Phone)
508-647-7001 (Fax)
The MathWorks, Inc.
3 Apple Hill Drive
Natick, MA 01760-2098
For contact information about worldwide offices, see the MathWorks Web site.
Bioinformatics Toolbox Users Guide
COPYRIGHT 20032014 by The MathWorks, Inc.
The software described in this document is furnished under a license agreement. The software may be used
or copied only under the terms of the license agreement. No part of this manual may be photocopied or
reproduced in any form without prior written consent from The MathWorks, Inc.
FEDERAL ACQUISITION: This provision applies to all acquisitions of the Program and Documentation
by, for, or through the federal government of the United States. By accepting delivery of the Program
or Documentation, the government hereby agrees that this software or documentation qualifies as
commercial computer software or commercial computer software documentation as such terms are used
or defined in FAR 12.212, DFARS Part 227.72, and DFARS 252.227-7014. Accordingly, the terms and
conditions of this Agreement and only those rights specified in this Agreement, shall pertain to and govern
the use, modification, reproduction, release, performance, display, and disclosure of the Program and
Documentation by the federal government (or other entity acquiring for or through the federal government)
and shall supersede any conflicting contractual terms or conditions. If this License fails to meet the
governments needs or is inconsistent in any respect with federal procurement law, the government agrees
to return the Program and Documentation, unused, to The MathWorks, Inc.

Trademarks

MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See
www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand
names may be trademarks or registered trademarks of their respective holders.
Patents

MathWorks products are protected by one or more U.S. patents. Please see
www.mathworks.com/patents for more information.

Revision History

September 2003
June 2004
November 2004
March 2005
May 2005
September 2005
November 2005
March 2006
May 2006
September 2006
March 2007
April 2007
September 2007
March 2008
October 2008
March 2009
September 2009
March 2010
September 2010
April 2011
September 2011
March 2012
September 2012
March 2013
September 2013
March 2014

Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online
Online

only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only
only

New for Version 1.0 (Release 13SP1+)

Revised for Version 1.1 (Release 14)
Revised for Version 2.0 (Release 14SP1+)
Revised for Version 2.0.1 (Release 14SP2)
Revised for Version 2.1 (Release 14SP2+)
Revised for Version 2.1.1 (Release 14SP3)
Revised for Version 2.2 (Release 14SP3+)
Revised for Version 2.2.1 (Release 2006a)
Revised for Version 2.3 (Release 2006a+)
Revised for Version 2.4 (Release 2006b)
Revised for Version 2.5 (Release 2007a)
Revised for Version 2.6 (Release 2007a+)
Revised for Version 3.0 (Release 2007b)
Revised for Version 3.1 (Release 2008a)
Revised for Version 3.2 (Release 2008b)
Revised for Version 3.3 (Release 2009a)
Revised for Version 3.4 (Release 2009b)
Revised for Version 3.5 (Release 2010a)
Revised for Version 3.6 (Release 2010b)
Revised for Version 3.7 (Release 2011a)
Revised for Version 4.0 (Release 2011b)
Revised for Version 4.1 (Release 2012a)
Revised for Version 4.2 (Release 2012b)
Revised for Version 4.3 (Release 2013a)
Revised for Version 4.3.1 (Release 2013b)
Revised for Version 4.4 (Release 2014a)

Contents
Getting Started

1
Bioinformatics Toolbox Product Description . . . . . . . . .
Key Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1-2
1-2

Product Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Expected Users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1-3
1-3
1-5

Installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Installing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Required Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Optional Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1-6
1-6
1-6
1-6

Features and Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Data Formats and Databases . . . . . . . . . . . . . . . . . . . . . . . .
Sequence Alignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sequence Utilities and Statistics . . . . . . . . . . . . . . . . . . . . .
Protein Property Analysis . . . . . . . . . . . . . . . . . . . . . . . . . .
Phylogenetic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Microarray Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . .
Microarray Data Storage . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mass Spectrometry Data Analysis . . . . . . . . . . . . . . . . . . . .
Graph Theory Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Graph Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Statistical Learning and Visualization . . . . . . . . . . . . . . . .
Prototyping and Development Environment . . . . . . . . . . . .
Data Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Algorithm Sharing and Application Deployment . . . . . . . .

1-9
1-9
1-11
1-11
1-12
1-13
1-13
1-14
1-15
1-18
1-19
1-19
1-20
1-20
1-21

Exchange Bioinformatic Data Between Excel and

MATLAB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Using Excel and MATLAB Together . . . . . . . . . . . . . . . . . .
About the Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Before Running the Example . . . . . . . . . . . . . . . . . . . . . . . .
Running the Example for the Entire Data Set . . . . . . . . . .

1-22
1-22
1-22
1-23
1-23

Editing Formulas to Run the Example on a Subset of the

Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Using the Spreadsheet Link EX Interface to Interact With
the Data in MATLAB . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Get Information from Web Database . . . . . . . . . . . . . . . .
What Are get Functions? . . . . . . . . . . . . . . . . . . . . . . . . . . .
Creating the getpubmed Function . . . . . . . . . . . . . . . . . . . .

1-27
1-28
1-31
1-31
1-32

High-Throughput Sequence Analysis

2
Work with Large Multi-Entry Text Files . . . . . . . . . . . . .
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
What Files Can You Access? . . . . . . . . . . . . . . . . . . . . . . . . .
Before You Begin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Create a BioIndexedFile Object to Access Your Source
File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Determine the Number of Entries Indexed By a
BioIndexedFile Object . . . . . . . . . . . . . . . . . . . . . . . . . . .
Retrieve Entries from Your Source File . . . . . . . . . . . . . . . .
Read Entries from Your Source File . . . . . . . . . . . . . . . . . .

2-2
2-2
2-2
2-3

Manage Short-Read Sequence Data in Objects . . . . . . .

Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Represent Sequence and Quality Data in a BioRead
Object . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Represent Sequence, Quality, and Alignment/Mapping
Data in a BioMap Object . . . . . . . . . . . . . . . . . . . . . . . . .
Retrieve Information from a BioRead or BioMap Object . .
Set Information in a BioRead or BioMap Object . . . . . . . . .
Determine Coverage of a Reference Sequence . . . . . . . . . . .
Construct Sequence Alignments to a Reference
Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Filter Read Sequences Using SAM Flags . . . . . . . . . . . . . .

2-8
2-8

Store and Manage Feature Annotations in Objects . . . .

Represent Feature Annotations in a GFFAnnotation or
GTFAnnotation Object . . . . . . . . . . . . . . . . . . . . . . . . . . .

vi

Contents

2-4
2-5
2-5
2-6

2-9
2-11
2-15
2-17
2-18
2-20
2-21
2-23
2-23

Construct an Annotation Object . . . . . . . . . . . . . . . . . . . . . .

Retrieve General Information from an Annotation
Object . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Access Data in an Annotation Object . . . . . . . . . . . . . . . . . .
Use Feature Annotations with Short-Read Sequence
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2-23

Visualize and Investigate Short-Read Alignments . . . .

When to Use the NGS Browser to Visualize and Investigate
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Open the NGS Browser . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Import Data into the NGS Browser . . . . . . . . . . . . . . . . . . .
Zoom and Pan to a Specific Region of the Alignment . . . . .
View Coverage of the Reference Sequence . . . . . . . . . . . . . .
View the Pileup View of Short Reads . . . . . . . . . . . . . . . . . .
Compare Alignments of Multiple Data Sets . . . . . . . . . . . .
View Location, Quality Scores, and Mapping
Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Flag Reads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Evaluate and Flag Mismatches . . . . . . . . . . . . . . . . . . . . . .
View Insertions and Deletions . . . . . . . . . . . . . . . . . . . . . . .
View Feature Annotations . . . . . . . . . . . . . . . . . . . . . . . . . .
Print and Export the Browser Image . . . . . . . . . . . . . . . . . .

2-31

2-24
2-25
2-26

2-31
2-32
2-33
2-36
2-37
2-38
2-39
2-40
2-41
2-42
2-43
2-43
2-43

Identifying Differentially Expressed Genes from

RNA-Seq Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2-44

Exploring Protein-DNA Binding Sites from Paired-End

ChIP-Seq Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2-61

Exploring Genome-wide Differences in DNA

Methylation Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2-78

Sequence Analysis

3
Exploring a Nucleotide Sequence Using Command
Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3-2
3-2

vii

viii

Contents

Searching the Web for Sequence Information . . . . . . . . . . .

Reading Sequence Information from the Web . . . . . . . . . . .
Determining Nucleotide Composition . . . . . . . . . . . . . . . . .
Determining Codon Composition . . . . . . . . . . . . . . . . . . . . .
Open Reading Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Amino Acid Conversion and Composition . . . . . . . . . . . . . .

3-2
3-5
3-6
3-11
3-15
3-18

Exploring a Nucleotide Sequence Using the Sequence

Viewer App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of the Sequence Viewer . . . . . . . . . . . . . . . . . . . .
Importing a Sequence into the Sequence Viewer . . . . . . . .
Viewing Nucleotide Sequence Information . . . . . . . . . . . . .
Searching for Words . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exploring Open Reading Frames . . . . . . . . . . . . . . . . . . . . .
Closing the Sequence Viewer . . . . . . . . . . . . . . . . . . . . . . . .

3-22
3-22
3-22
3-24
3-26
3-29
3-32

Explore a Protein Sequence Using the Sequence Viewer

App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of the Sequence Viewer . . . . . . . . . . . . . . . . . . . .
Viewing Amino Acid Sequence Statistics . . . . . . . . . . . . . . .
Closing the Sequence Viewer . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3-33
3-33
3-33
3-37
3-37

Sequence Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Find a Model Organism to Study . . . . . . . . . . . . . . . . . . . . .
Retrieve Sequence Information from a Public Database . .
Search a Public Database for Related Genes . . . . . . . . . . .
Locate Protein Coding Sequences . . . . . . . . . . . . . . . . . . . .
Compare Amino Acid Sequences . . . . . . . . . . . . . . . . . . . . .

3-38
3-38
3-38
3-41
3-43
3-45
3-49

View and Align Multiple Sequences . . . . . . . . . . . . . . . . .

Overview of the Sequence Alignment and Phylogenetic Tree
Apps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Load Sequence Data and Viewing the Phylogenetic Tree . .
Select a Subset of Data from the Phylogenetic Tree . . . . . .
Align Multiple Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . .
Adjust Multiple Sequence Alignments Manually . . . . . . . .
Close the Sequence Alignment App . . . . . . . . . . . . . . . . . . .

3-58
3-58
3-58
3-59
3-61
3-62
3-65

Microarray Analysis

4
Managing Gene Expression Data in Objects . . . . . . . . . .

4-2

Representing Expression Data Values in DataMatrix

Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of DataMatrix Objects . . . . . . . . . . . . . . . . . . . . .
Constructing DataMatrix Objects . . . . . . . . . . . . . . . . . . . .
Getting and Setting Properties of a DataMatrix Object . . .
Accessing Data in DataMatrix Objects . . . . . . . . . . . . . . . .

4-5
4-5
4-6
4-7
4-8

Representing Expression Data Values in ExptData

Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of ExptData Objects . . . . . . . . . . . . . . . . . . . . . . .
Constructing ExptData Objects . . . . . . . . . . . . . . . . . . . . . .
Using Properties of an ExptData Object . . . . . . . . . . . . . . .
Using Methods of an ExptData Object . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4-11
4-11
4-12
4-12
4-13
4-14

Representing Sample and Feature Metadata in

MetaData Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of MetaData Objects . . . . . . . . . . . . . . . . . . . . . . .
Constructing MetaData Objects . . . . . . . . . . . . . . . . . . . . . .
Using Properties of a MetaData Object . . . . . . . . . . . . . . . .
Using Methods of a MetaData Object . . . . . . . . . . . . . . . . .

4-15
4-15
4-16
4-19
4-20

Representing Experiment Information in a MIAME

Object . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of MIAME Objects . . . . . . . . . . . . . . . . . . . . . . . .
Constructing MIAME Objects . . . . . . . . . . . . . . . . . . . . . . .
Using Properties of a MIAME Object . . . . . . . . . . . . . . . . . .
Using Methods of a MIAME Object . . . . . . . . . . . . . . . . . . .

4-22
4-22
4-22
4-25
4-25

Representing All Data in an ExpressionSet Object . . . .

Overview of ExpressionSet Objects . . . . . . . . . . . . . . . . . . .
Constructing ExpressionSet Objects . . . . . . . . . . . . . . . . . .
Using Properties of an ExpressionSet Object . . . . . . . . . . .
Using Methods of an ExpressionSet Object . . . . . . . . . . . . .

4-27
4-27
4-29
4-30
4-30

ix

Visualizing Microarray Images . . . . . . . . . . . . . . . . . . . . .

Overview of the Mouse Example . . . . . . . . . . . . . . . . . . . . .
Exploring the Microarray Data Set . . . . . . . . . . . . . . . . . . .
Spatial Images of Microarray Data . . . . . . . . . . . . . . . . . . .
Statistics of the Microarrays . . . . . . . . . . . . . . . . . . . . . . . .
Scatter Plots of Microarray Data . . . . . . . . . . . . . . . . . . . . .

4-33
4-33
4-34
4-36
4-46
4-48

Analyzing Gene Expression Profiles . . . . . . . . . . . . . . . . .

Overview of the Yeast Example . . . . . . . . . . . . . . . . . . . . . .
Exploring the Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Filtering Genes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Clustering Genes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Principal Component Analysis . . . . . . . . . . . . . . . . . . . . . . .

4-57
4-57
4-57
4-61
4-64
4-68

Detecting DNA Copy Number Alteration in Array-Based

CGH Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-72
Exploring Gene Expression Data . . . . . . . . . . . . . . . . . . . .

4-85

Phylogenetic Analysis

Contents

Overview of Phylogenetic Analysis . . . . . . . . . . . . . . . . . .

5-2

Building a Phylogenetic Tree . . . . . . . . . . . . . . . . . . . . . . .

Overview of the Primate Example . . . . . . . . . . . . . . . . . . . .
Searching NCBI for Phylogenetic Data . . . . . . . . . . . . . . . .
Creating a Phylogenetic Tree for Five Species . . . . . . . . . .
Creating a Phylogenetic Tree for Twelve Species . . . . . . . .
Exploring the Phylogenetic Tree . . . . . . . . . . . . . . . . . . . . .

5-3
5-3
5-5
5-6
5-9
5-11

Phylogenetic Tree App Reference . . . . . . . . . . . . . . . . . . .

Overview of the Phylogenetic Tree App . . . . . . . . . . . . . . . .
Opening the Phylogenetic Tree App . . . . . . . . . . . . . . . . . . .
File Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Tools Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Window Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Help Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5-16
5-16
5-16
5-18
5-29
5-38
5-38

1
Getting Started
Bioinformatics Toolbox Product Description on page 1-2
Product Overview on page 1-3
Installation on page 1-6
Features and Functions on page 1-9
Exchange Bioinformatic Data Between Excel and MATLAB on page 1-22
Get Information from Web Database on page 1-31

Getting Started

Bioinformatics Toolbox Product Description

Read, analyze, and visualize genomic and proteomic data
Bioinformatics Toolbox provides algorithms and apps for Next Generation
Sequencing (NGS), microarray analysis, mass spectrometry, and gene
ontology. Using toolbox functions, you can read genomic and proteomic
data from standard file formats such as SAM, FASTA, CEL, and CDF, as
well as from online databases such as the NCBI Gene Expression Omnibus
and GenBank. You can explore and visualize this data with sequence
browsers, spatial heatmaps, and clustergrams. The toolbox also provides
statistical techniques for detecting peaks, imputing values for missing data,
and selecting features.
You can combine toolbox functions to support common bioinformatics
workflows. You can use ChIP-Seq data to identify transcription factors;
analyze RNA-Seq data to identify differentially expressed genes; identify copy
number variants and SNPs in microarray data; and classify protein profiles
using mass spectrometry data.

Key Features
Next Generation Sequencing analysis and browser
Sequence analysis and visualization, including pairwise and multiple
sequence alignment and peak detection
Microarray data analysis, including reading, filtering, normalizing, and
visualization
Mass spectrometry analysis, including preprocessing, classification, and
marker identification
Phylogenetic tree analysis
Graph theory functions, including interaction maps, hierarchy plots, and
pathways
Data import from genomic, proteomic, and gene expression files, including
SAM, FASTA, CEL, and CDF, and from databases such as NCBI and
GenBank

1-2

Product Overview

Product Overview
In this section...
Features on page 1-3
Expected Users on page 1-5

Features
The Bioinformatics Toolbox product extends the MATLAB environment
to provide an integrated software environment for genome and proteome
analysis. Scientists and engineers can answer questions, solve problems,
prototype new algorithms, and build applications for drug discovery and
design, genetic engineering, and biological research. An introduction to these
features will help you to develop a conceptual model for working with the
toolbox and your biological data.
The Bioinformatics Toolbox product includes many functions to help you
with genome and proteome analysis. Most functions are implemented in the
MATLAB programming language, with the source available for you to view.
This open environment lets you explore and customize the existing toolbox
algorithms or develop your own.
You can use the basic bioinformatic functions provided with this toolbox
to create more complex algorithms and applications. These robust and
well-tested functions are the functions that you would otherwise have to
create yourself.
Toolbox features and functions fall within these categories:
Data formats and databases Connect to Web-accessible databases
containing genomic and proteomic data. Read and convert between
multiple data formats.
High-throughput sequencing Gene expression and transcription
factor analysis of next-generation sequencing data, including RNA-Seq
and ChIP-Seq.
Sequence analysis Determine the statistical characteristics of a
sequence, align two sequences, and multiply align several sequences.

1-3

Getting Started

Model patterns in biological sequences using hidden Markov model (HMM)

profiles.
Phylogenetic analysis Create and manipulate phylogenetic tree data.
Microarray data analysis Read, normalize, and visualize microarray
data.
Mass spectrometry data analysis Analyze and enhance raw mass
spectrometry data.
Statistical learning Classify and identify features in data sets with
statistical learning tools.
Programming interface Use other bioinformatic software (BioPerl
and BioJava) within the MATLAB environment.
The field of bioinformatics is rapidly growing and will become increasingly
important as biology becomes a more analytical science. The toolbox provides
an open environment that you can customize for development and deployment
of the analytical tools you will need.
Prototype and develop algorithms Prototype new ideas in an open
and extensible environment. Develop algorithms using efficient string
processing and statistical functions, view the source code for existing
functions, and use the code as a template for customizing, improving,
or creating your own functions. See Prototyping and Development
Environment on page 1-20.
Visualize data Visualize sequences and alignments, gene expression
data, phylogenetic trees, mass spectrometry data, protein structure,
and relationships between data with interconnected graphs. See Data
Visualization on page 1-20.
Share and deploy applications Use an interactive GUI builder to
develop a custom graphical front end for your data analysis programs.
Create standalone applications that run separately from the MATLAB
environment. See Algorithm Sharing and Application Deployment on
page 1-21.

1-4

Product Overview

Expected Users
The Bioinformatics Toolbox product is intended for computational biologists
and research scientists who need to develop new algorithms or implement
published ones, visualize results, and create standalone applications.
Industry/Professional Increasingly, drug discovery methods are being
supported by engineering practice. This toolbox supports tool builders
who want to create applications for the biotechnology and pharmaceutical
industries.
Education/Professor/Student This toolbox is well suited for learning
and teaching genome and proteome analysis techniques. Educators
and students can concentrate on bioinformatic algorithms instead of
programming basic functions such as reading and writing to files.
While the toolbox includes many bioinformatic functions, it is not intended
to be a complete set of tools for scientists to analyze their biological data.
However, the MATLAB environment is ideal for rapidly designing and
prototyping the tools you need.

1-5

Getting Started

Installation
In this section...
Installing on page 1-6
Required Software on page 1-6
Optional Software on page 1-6

Installing
Install the Bioinformatics Toolbox software from a DVD or Web release
using the MathWorks Installer. For more information, see the installation
documentation.

Required Software
The Bioinformatics Toolbox software requires the following MathWorks
products to be installed on your computer.
Required Software

Description

MATLAB

Provides a command-line interface and integrated

software environment for the Bioinformatics
Toolbox software.
Bioinformatics Toolbox software requires the
current version ofMATLAB.

Statistics Toolbox

Provides basic statistics and probability functions

used by the Bioinformatics Toolbox software.
Bioinformatics Toolbox software requires the
current version ofStatistics Toolbox.

Optional Software
MATLAB and the Bioinformatics Toolbox software environment is open and
extensible. In this environment you can interactively explore ideas, prototype
new algorithms, and develop complete solutions to problems in bioinformatics.
MATLAB facilitates computation, visualization, prototyping, and deployment.

1-6

Installation

Using the Bioinformatics Toolbox software with other MATLAB toolboxes and
products will allow you to do advanced algorithm development and solve
multidisciplinary problems.
Optional Software

Description

Parallel Computing
Toolbox

Perform parallel bioinformatic computations on

multicore computers and computer clusters. For
an example of batch processing through parallel
computing, see the Batch Processing of Spectra
Using Distributed Computing.

Signal Processing
Toolbox

Process signal data from bioanalytical

instrumentation. Examples include acquisition
of fluorescence data for DNA sequence analyzers,
fluorescence data for microarray scanners, and
mass spectrometric data from protein analyses.

Image Processing
Toolbox

Create complex and custom image processing

algorithms for data from microarray scanners.

SimBiology

Model, simulate, and analyze biochemical systems.

Optimization
Toolbox

Use nonlinear optimization to predict the

secondary structure of proteins and the structure
of other biological macromolecules.

Neural Network
Toolbox

Use neural networks to solve problems where

algorithms are not available. For example, you
can train neural networks for pattern recognition
using large sets of sequence data.

Database Toolbox

Create your own in-house databases for sequence

data with custom annotations.

MATLAB
Compiler

Create standalone applications from MATLAB

GUI applications, and create dynamic link
libraries from MATLAB functions to use with any
programming environment.

MATLAB Builder
NE

Create COM objects to use with any COM-based

programming environment.

1-7

1-8

Getting Started

Optional Software

Description

MATLAB Builder JA

Integrate MATLAB applications into your

organizations Java programs by creating a Java
wrapper around the application.

MATLAB Builder EX

Create Microsoft Excel add-in functions

from MATLAB functions to use with Excel
spreadsheets.

Spreadsheet Link
EX

Connect Microsoft Excel with the MATLAB

Workspace to exchange data and to use MATLAB
computational and visualization functions. For
more information, see Exchange Bioinformatic
Data Between Excel and MATLAB on page 1-22.

Features and Functions

Features and Functions

In this section...
Data Formats and Databases on page 1-9
Sequence Alignments on page 1-11
Sequence Utilities and Statistics on page 1-11
Protein Property Analysis on page 1-12
Phylogenetic Analysis on page 1-13
Microarray Data Analysis on page 1-13
Microarray Data Storage on page 1-14
Mass Spectrometry Data Analysis on page 1-15
Graph Theory Functions on page 1-18
Graph Visualization on page 1-19
Statistical Learning and Visualization on page 1-19
Prototyping and Development Environment on page 1-20
Data Visualization on page 1-20
Algorithm Sharing and Application Deployment on page 1-21

Data Formats and Databases

The toolbox accesses many of the databases on the Web and other online data
sources. It allows you to copy data into the MATLAB Workspace, and read
and write to files with standard bioinformatic formats. It also reads many
common genome file formats, so that you do not have to write and maintain
your own file readers.
Web-based databases You can directly access public databases on the
Web and copy sequence and gene expression information into the MATLAB
environment.
The sequence databases currently supported are GenBank (getgenbank),
GenPept (getgenpept), European Molecular Biology Laboratory (EMBL)
(getembl), and Protein Data Bank (PDB) (getpdb). You can also access data

1-9

Getting Started

from the NCBI Gene Expression Omnibus (GEO) Web site by using a single
function (getgeodata).
Get multiply aligned sequences (gethmmalignment), hidden Markov model
profiles (gethmmprof), and phylogenetic tree data (gethmmtree) from the
PFAM database.
Gene Ontology database Load the database from the Web into
a gene ontology object (geneont.geneont). Select sections of the
ontology with methods for the geneont object (geneont.getancestors,
geneont.getdescendants, geneont.getmatrix, geneont.getrelatives),
and manipulate data with utility functions (goannotread, num2goid).
Read data from instruments Read data generated from gene
sequencing instruments (scfread, joinseq, traceplot), mass spectrometers
(jcampread), and Agilent microarray scanners (agferead).
Reading data formats The toolbox provides a number of functions for
reading data from common bioinformatic file formats.
Sequence data: GenBank (genbankread), GenPept (genpeptread), EMBL
(emblread), PDB (pdbread), and FASTA (fastaread)
Multiply aligned sequences: ClustalW and GCG formats (multialignread)
Gene expression data from microarrays: Gene Expression Omnibus (GEO)
data (geosoftread), GenePix data in GPR and GAL files (gprread,
galread), SPOT data (sptread), Affymetrix GeneChip data (affyread),
and ImaGene results files (imageneread)
Hidden Markov model profiles: PFAM-HMM file (pfamhmmread)
Writing data formats The functions for getting data from the Web
include the option to save the data to a file. However, there is a function to
write data to a file using the FASTA format (fastawrite).
BLAST searches Request Web-based BLAST searches (blastncbi), get
the results from a search (getblast) and read results from a previously saved
BLAST formatted report file (blastread).

1-10

Features and Functions

The MATLAB environment has built-in support for other industry-standard

file formats including Microsoft Excel and comma-separated-value (CSV) files.
Additional functions perform ASCII and low-level binary I/O, allowing you to
develop custom functions for working with any data format.

Sequence Alignments
You can select from a list of analysis methods to compare nucleotide or amino
acid sequences using pairwise or multiple sequence alignment functions.
Pairwise sequence alignment Efficient implementations of standard
algorithms such as the Needleman-Wunsch (nwalign) and Smith-Waterman
(swalign) algorithms for pairwise sequence alignment. The toolbox also
includes standard scoring matrices such as the PAM and BLOSUM
families of matrices (blosum, dayhoff, gonnet, nuc44, pam). Visualize
sequence similarities with seqdotplot and sequence alignment results with
showalignment.
Multiple sequence alignment Functions for multiple sequence
alignment (multialign, profalign) and functions that support multiple
sequences (multialignread, fastaread, showalignment). There is also a
graphical interface (seqalignviewer) for viewing the results of a multiple
sequence alignment and manually making adjustment.
Multiple sequence profiles Implementations for multiple alignment and
profile hidden Markov model algorithms (gethmmprof, gethmmalignment,
gethmmtree, pfamhmmread, hmmprofalign, hmmprofestimate,
hmmprofgenerate, hmmprofmerge, hmmprofstruct, showhmmprof).
Biological codes Look up the letters or numeric equivalents for
commonly used biological codes (aminolookup, baselookup, geneticcode,
revgeneticcode).

Sequence Utilities and Statistics

You can manipulate and analyze your sequences to gain a deeper
understanding of the physical, chemical, and biological characteristics of
your data. Use a graphical user interface (GUI) with many of the sequence
functions in the toolbox (seqviewer).

1-11

Getting Started

Sequence conversion and manipulation The toolbox provides routines

for common operations, such as converting DNA or RNA sequences to amino
acid sequences, that are basic to working with nucleic acid and protein
sequences (aa2int, aa2nt, dna2rna, rna2dna, int2aa, int2nt, nt2aa, nt2int,
seqcomplement, seqrcomplement, seqreverse).
You can manipulate your sequence by performing an in silico digestion with
restriction endonucleases (restrict) and proteases (cleave).
Sequence statistics Determine various statistics about a sequence
(aacount, basecount, codoncount, dimercount, nmercount, ntdensity,
codonbias, cpgisland, oligoprop), search for specific patterns within a
sequence (seqshowwords, seqwordcount), or search for open reading frames
(seqshoworfs). In addition, you can create random sequences for test cases
(randseq).
Sequence utilities Determine a consensus sequence from a set of multiply
aligned amino acid, nucleotide sequences (seqconsensus, or a sequence
profile (seqprofile). Format a sequence for display (seqdisp) or graphically
show a sequence alignment with frequency data (seqlogo).
Additional MATLAB functions efficiently handle string operations with
regular expressions (regexp, seq2regexp) to look for specific patterns in a
sequence and search through a library for string matches (seqmatch).
Look for possible cleavage sites in a DNA/RNA sequence by searching for
palindromes (palindromes).

Protein Property Analysis

You can use a collection of protein analysis methods to extract information
from your data. You can determine protein characteristics and simulate
enzyme cleavage reactions. The toolbox provides functions to calculate various
properties of a protein sequence, such as the atomic composition (atomiccomp),
molecular weight (molweight), and isoelectric point (isoelectric). You can
cleave a protein with an enzyme (cleave, rebasecuts) and create distance
and Ramachandran plots for PDB data (pdbdistplot, ramachandran). The
toolbox contains a graphical user interface for protein analysis (proteinplot)
and plotting 3-D protein and other molecular structures with information
from molecule model files, such as PDB files (molviewer).

1-12

Features and Functions

Amino acid sequence utilities Calculate amino acid statistics for a

sequence (aacount) and get information about character codes (aminolookup).

Phylogenetic Analysis
You can use functions for phylogenetic tree building and analysis. There is
also a GUI to draw phylograms (trees).
Phylogenetic tree data Read and write Newick-formatted tree files
(phytreeread, phytreewrite) into the MATLAB Workspace as phylogenetic
tree objects (phytree).
Create a phylogenetic tree Calculate the pairwise distance between
biological sequences (seqpdist), estimate the substitution rates (dnds,
dndsml), build a phylogenetic tree from pairwise distances (seqlinkage,
seqneighjoin, reroot), and view the tree in an interactive GUI that allows
you to view, edit, and explore the data (phytreeviewer or view). This GUI
also allows you to prune branches, reorder, rename, and explore distances.
Phylogenetic tree object methods You can access the functionality
of the phytreeviewer GUI using methods for a phylogenetic tree object
(phytree). Get property values (get) and node names (getbyname). Calculate
the patristic distances between pairs of leaf nodes (pdist, weights)
and draw a phylogenetic tree object in a MATLAB Figure window as a
phylogram, cladogram, or radial treeplot (plot). Manipulate tree data by
selecting branches and leaves using a specified criterion (select, subtree)
and removing nodes (prune). Compare trees (getcanonical) and use
Newick-formatted strings (getnewickstr).

Microarray Data Analysis

The MATLAB environment is widely used for microarray data analysis,
including reading, filtering, normalizing, and visualizing microarray data.
However, the standard normalization and visualization tools that scientists
use can be difficult to implement. The toolbox includes these standard
functions:
Microarray data Read Affymetrix GeneChip files (affyread) and plot
data (probesetplot), ImaGene results files (imageneread), SPOT files
(sptread) and Agilent microarray scanner files (agferead). Read GenePix

1-13

Getting Started

GPR files (gprread) and GAL files (galread). Get Gene Expression Omnibus
(GEO) data from the Web (getgeodata) and read GEO data from files
(geosoftread).
A utility function (magetfield) extracts data from one of the microarray
reader functions (gprread, agferead, sptread, imageneread).
Microarray normalization and filtering The toolbox provides a
number of methods for normalizing microarray data, such as lowess
normalization (malowess) and mean normalization (manorm), or across
multiple arrays (quantilenorm). You can use filtering functions to
clean raw data before analysis (geneentropyfilter, genelowvalfilter,
generangefilter, genevarfilter), and calculate the range and variance of
values (exprprofrange, exprprofvar).
Microarray visualization The toolbox contains routines for visualizing
microarray data. These routines include spatial plots of microarray data
(maimage, redgreencmap), box plots (maboxplot), loglog plots (maloglog),
and intensity-ratio plots (mairplot). You can also view clustered expression
profiles (clustergram, redgreencmap). You can create 2-D scatter plots of
principal components from the microarray data (mapcaplot).
Microarray utility functions Use the following functions to work
with Affymetrix GeneChip data sets. Get library information for a probe
(probelibraryinfo), gene information from a probe set (probesetlookup),
and probe set values from CEL and CDF information (probesetvalues).
Show probe set information from NetAffx Analysis Center (probesetlink)
and plot probe set values (probesetplot).
The toolbox accesses statistical routines to perform cluster analysis and
to visualize the results, and you can view your data through statistical
visualizations such as dendrograms, classification, and regression trees.

Microarray Data Storage

The toolbox includes functions, objects, and methods for creating, storing, and
accessing microarray data.
The object constructor function, DataMatrix, lets you create a DataMatrix
object to encapsulate data and metadata from a microarray experiment. A

1-14

Features and Functions

DataMatrix object stores experimental data in a matrix, with rows typically

corresponding to gene names or probe identifiers, and columns typically
corresponding to sample identifiers. A DataMatrix object also stores
metadata, including the gene names or probe identifiers (as the row names)
and sample identifiers (as the column names).
You can reference microarray expression values in a DataMatrix object the
same way you reference data in a MATLAB array, that is, by using linear or
logical indexing. Alternately, you can reference this experimental data by
gene (probe) identifiers and sample identifiers. Indexing by these identifiers
lets you quickly and conveniently access subsets of the data without having
to maintain additional index arrays.
Many MATLAB operators and arithmetic functions are available to
DataMatrix objects by means of methods. These methods let you modify,
combine, compare, analyze, plot, and access information from DataMatrix
objects. Additionally, you can easily extend the functionality by using
general element-wise functions, dmarrayfun and dmbsxfun, and by manually
accessing the properties of a DataMatrix object.
Note For more information on creating and using DataMatrix objects, see
Representing Expression Data Values in DataMatrix Objects on page 4-5.

Mass Spectrometry Data Analysis

The mass spectrometry functions preprocess and classify raw data from
SELDI-TOF and MALDI-TOF spectrometers and use statistical learning
functions to identify patterns.
Reading raw data Load raw mass/charge and ion intensity data from
comma-separated-value (CSV) files, or read a JCAMP-DX-formatted file with
mass spectrometry data (jcampread) into the MATLAB environment.
You can also have data in TXT files and use the importdata function.
Preprocessing raw data Resample high-resolution data to a lower
resolution (msresample) where the extra data points are not needed. Correct
the baseline (msbackadj). Align a spectrum to a set of reference masses

1-15

Getting Started

(msalign) and visually verify the alignment (msheatmap). Normalize the

area between spectra for comparing (msnorm), and filter out noise (mslowess
and mssgolay).
Spectrum analysis Load spectra into a GUI (msviewer) for selecting mass
peaks and further analysis.
The following graphic illustrates the roles of the various mass spectrometry
functions in the toolbox.

1-16

Features and Functions

mzXML File
mzxmlread

mzXML Structure
mzxml2peaks

Peak Lists
(Centroided Data)

mspeaks

Raw
Data

msdotplot

Plot

msheatmap

Plot

msppresample

Reconstructed
Data

Semicontinuous Signal

msviewer

Mass
Spectra
Viewer

msresample

1-17

Getting Started

Graph Theory Functions

Graph theory functions in the toolbox apply basic graph theory algorithms to
sparse matrices. A sparse matrix represents a graph, any nonzero entries in
the matrix represent the edges of the graph, and the values of these entries
represent the associated weight (cost, distance, length, or capacity) of the
edge. Graph algorithms that use the weight information will cancel the edge
if a NaN or an Inf is found. Graph algorithms that do not use the weight
information will consider the edge if a NaN or an Inf is found, because these
algorithms look only at the connectivity described by the sparse matrix and
not at the values stored in the sparse matrix.
Sparse matrices can represent four types of graphs:
Directed Graph Sparse matrix, either double real or logical. Row
(column) index indicates the source (target) of the edge. Self-loops (values
in the diagonal) are allowed, although most of the algorithms ignore these
values.
Undirected Graph Lower triangle of a sparse matrix, either double
real or logical. An algorithm expecting an undirected graph ignores values
stored in the upper triangle of the sparse matrix and values in the diagonal.
Direct Acyclic Graph (DAG) Sparse matrix, double real or logical,
with zero values in the diagonal. While a zero-valued diagonal is a
requirement of a DAG, it does not guarantee a DAG. An algorithm expecting
a DAG will not test for cycles because this will add unwanted complexity.
Spanning Tree Undirected graph with no cycles and with one
connected component.
There are no attributes attached to the graphs; sparse matrices representing
all four types of graphs can be passed to any graph algorithm. All functions
will return an error on nonsquare sparse matrices.
Graph algorithms do not pretest for graph properties because such tests
can introduce a time penalty. For example, there is an efficient shortest
path algorithm for DAG, however testing if a graph is acyclic is expensive
compared to the algorithm. Therefore, it is important to select a graph theory
function and properties appropriate for the type of the graph represented by
your input matrix. If the algorithm receives a graph type that differs from
what it expects, it will either:

1-18

Features and Functions

Return an error when it reaches an inconsistency. For example, if you pass

a cyclic graph to the graphshortestpath function and specify Acyclic as
the method property.
Produce an invalid result. For example, if you pass a directed graph to a
function with an algorithm that expects an undirected graph, it will ignore
values in the upper triangle of the sparse matrix.
The graph theory functions include graphallshortestpaths, graphconncomp,
graphisdag, graphisomorphism, graphisspantree, graphmaxflow,
graphminspantree, graphpred2path, graphshortestpath, graphtopoorder,
and graphtraverse.

Graph Visualization
The toolbox includes functions, objects, and methods for creating, viewing,
and manipulating graphs such as interactive maps, hierarchy plots, and
pathways. This allows you to view relationships between data.
The object constructor function (biograph) lets you create a biograph object to
hold graph data. Methods of the biograph object let you calculate the position
of nodes (dolayout), draw the graph (view), get handles to the nodes and
edges (getnodesbyid and getedgesbynodeid) to further query information,
and find relations between the nodes (getancestors, getdescendants,
and getrelatives). There are also methods that apply basic graph theory
algorithms to the biograph object.
Various properties of a biograph object let you programmatically change the
properties of the rendered graph. You can customize the node representation,
for example, drawing pie charts inside every node (CustomNodeDrawFcn). Or
you can associate your own callback functions to nodes and edges of the graph,
for example, opening a Web page with more information about the nodes
(NodeCallback and EdgeCallback).

Statistical Learning and Visualization

You can classify and identify features in data sets, set up cross-validation
experiments, and compare different classification methods.

1-19

Getting Started

The toolbox provides functions that build on the classification and statistical
learning tools in the Statistics Toolbox software (classify, kmeans, and
treefit).
These functions include imputation tools (knnimpute), and K-nearest neighbor
classifiers (knnclassify).
Other functions include set up of cross-validation experiments (crossvalind)
and comparison of the performance of different classification methods
(classperf). In addition, there are tools for selecting diversity and
discriminating features (rankfeatures, randfeatures).

Prototyping and Development Environment

The MATLAB environment lets you prototype and develop algorithms and
easily compare alternatives.
Integrated environment Explore biological data in an environment
that integrates programming and visualization. Create reports and plots
with the built-in functions for mathematics, graphics, and statistics.
Open environment Access the source code for the toolbox functions.
The toolbox includes many of the basic bioinformatics functions you will
need to use, and it includes prototypes for some of the more advanced
functions. Modify these functions to create your own custom solutions.
Interactive programming language Test your ideas by typing
functions that are interpreted interactively with a language whose basic
data element is an array. The arrays do not require dimensioning and
allow you to solve many technical computing problems,
Using matrices for sequences or groups of sequences allows you to work
efficiently and not worry about writing loops or other programming controls.
Programming tools Use a visual debugger for algorithm development
and refinement and an algorithm performance profiler to accelerate
development.

Data Visualization
You can visually compare pairwise sequence alignments, multiply aligned
sequences, gene expression data from microarrays, and plot nucleic acid and

1-20

Features and Functions

protein characteristics. The 2-D and volume visualization features let you
create custom graphical representations of multidimensional data sets. You
can also create montages and overlays, and export finished graphics to an
Adobe PostScript image file or copy directly into Microsoft PowerPoint.

Algorithm Sharing and Application Deployment

The open MATLAB environment lets you share your analysis solutions with
other users, and it includes tools to create custom software applications.
With the addition of MATLAB Compiler software, you can create standalone
applications independent of the MATLAB environment, and, with the addition
of MATLAB Builder NE software, you can create GUIs and standalone
applications within other programming environments.
Share algorithms with other users You can share data analysis
algorithms created in the MATLAB language across all supported
platforms by giving files to other users. You can also create GUIs within the
MATLAB environment using the Graphical User Interface Development
Environment (GUIDE).
Deploy MATLAB GUIs Create a GUI within the MATLAB
environment using GUIDE, and then use MATLAB Compiler software
to create a standalone GUI application that runs separately from the
MATLAB environment.
Create dynamic link libraries (DLLs) Use MATLAB Compiler
software to create DLLs for your functions, and then link these libraries to
other programming environments such as C and C++.
Create COM objects Use MATLAB Builder NE software to create
COM objects, and then use a COM-compatible programming environment
(Visual Basic) to create a standalone application.
Create Excel add-ins Use MATLAB Builder EX software to
create Excel add-in functions, and then use these functions with Excel
spreadsheets.
Create Java classes Use MATLAB Builder JA software to
automatically generate Java classes from algorithms written in the
MATLAB programming language. You can run these classes outside the
MATLAB environment.

1-21

Getting Started

Exchange Bioinformatic Data Between Excel and MATLAB

In this section...
Using Excel and MATLAB Together on page 1-22
About the Example on page 1-22
Before Running the Example on page 1-23
Running the Example for the Entire Data Set on page 1-23
Editing Formulas to Run the Example on a Subset of the Data on page
1-27
Using the Spreadsheet Link EX Interface to Interact With the Data in
MATLAB on page 1-28

Using Excel and MATLAB Together

If you have bioinformatic data in an Excel (2007 or 2010) spreadsheet, use
Spreadsheet Link EX to:
Connect Excel with the MATLAB Workspace to exchange data
Use MATLAB and Bioinformatics Toolbox computational and visualization
functions

About the Example

Note The following example assumes you have Spreadsheet Link EX
software installed on your system.
The Excel file used in the following example contains data from DeRisi, J.L.,
Iyer, V.R., and Brown, P.O. (Oct. 24, 1997). Exploring the metabolic and
genetic control of gene expression on a genomic scale. Science 278(5338),
680686. PMID: 9381177. The data was filtered using the steps described
in Gene Expression Profile Analysis.

1-22

Exchange Bioinformatic Data Between Excel and MATLAB

Before Running the Example

1 If not already done, modify your system path to include the MATLAB root

folder as described in the Spreadsheet Link EX documentation.

2 If not already done, enable the Spreadsheet Link EX Add-In as described in

Customization in the Spreadsheet Link EX documentation.

3 Close MATLAB and Excel if they are open.
4 Start Excel 2007 or 2010 software. MATLAB and Spreadsheet Link EX

software automatically start.

5 From Excel, open the following file provided with the Bioinformatics

Toolbox software:
matlabroot\toolbox\bioinfo\biodemos\Filtered_Yeastdata.xlsm

Note matlabroot is the MATLAB root folder, which is where MATLAB

software is installed on your system.
6 In the Excel software, enable macros. Click the Developer tab, and then

select Macro Security from the Code group. (If the Developer tab is not
displayed on the Excel ribbon, consult Excel Help to display it.)

Running the Example for the Entire Data Set

1 In the provided Excel file, note that columns A through H contain data

from DeRisi et al. Also note that cells J5, J6, J7, and J12 contain formulas
using Spreadsheet Link EX functions MLPutMatrix and MLEvalString.
Tip To view a cells formula, select the cell, and then view the formula in
the formula bar

at the top of the Excel window.

2 Execute the formulas in cells J5, J6, J7, and J12, by selecting the cell,

pressing F2, and then pressing Enter.

1-23

Getting Started

Each of the first three cells contains a formula using the Spreadsheet Link
EX function MLPutMatrix, which creates a MATLAB variable from the
data in the spreadsheet. Cell J12 contains a formula using the Spreadsheet
Link EX function MLEvalString, which runs the Bioinformatics Toolbox
clustergram function using the three variables as input. For more
information on adding formulas using Spreadsheet Link EX functions,
see Enter Functions into Worksheet Cells in the Spreadsheet Link EX
documentation.

1-24

Exchange Bioinformatic Data Between Excel and MATLAB

Cells J5, J6, and J7 contain formulas

that use the MLPutMatrix function
to create three MATLAB variables.

Cell J12 contains a formula

that uses the MLEvalString function
to run the Bioinformatics Toolbox function
clustergram.

Cell J17 contains a formula

that uses a macro function,
Clustergram, created in
Visual Basic Editor.
3 Note that cell J17 contains a formula using a macro function Clustergram,

which was created in the Visual Basic Editor. Running this macro does the
same as the formulas in cells J5, J6, J7, and J12. Optionally, view the
Clustergram macro function by clicking the Developer tab, and then
clicking the Visual Basic button
. (If the Developer tab is not on the
Excel ribbon, consult Excel Help to display it.)

1-25

Getting Started

For more information on creating macros using Visual Basic Editor, see
Use Spreadsheet Link EX Functions in Macros in the Spreadsheet Link
EX documentation.
4 Execute the formula in cell J17 to analyze and visualize the data:
a Select cell J17.
b Press F2.
c Press Enter.

The macro function Clustergram runs creating three MATLAB variables

(data, Genes, and TimeSteps) and displaying a Clustergram window
containing dendrograms and a heat map of the data.

1-26

Exchange Bioinformatic Data Between Excel and MATLAB

Editing Formulas to Run the Example on a Subset

of the Data
1 Edit the formulas in cells J5 and J6 to analyze a subset of the data. Do

this by editing the formulas cell ranges to include data for only the first
30 genes:
a Select cell J5, and then press F2 to display the formula for editing.

Change H617 to H33, and then press Enter.

b Select cell J6, then press F2 to display the formula for editing. Change

A617 to A33, and then press Enter.

2 Run the formulas in cells J5, J6, J7, and J12 to analyze and visualize

a subset of the data:

a Select cell J5, press F2, and then press Enter.
b Select cell J6, press F2, and then press Enter.
c Select cell J7, press F2, and then press Enter.
d Select cell J12, press F2, and then press Enter.

1-27

Getting Started

Using the Spreadsheet Link EX Interface to Interact

With the Data in MATLAB
Use the MATLAB group on the right side of the Home tab to interact with
the data:

1-28

Exchange Bioinformatic Data Between Excel and MATLAB

For example, create a variable in MATLAB containing a 3-by-7 matrix of

the data, plot the data in a Figure window, and then add the plot to your
spreadsheet:
1 Click-drag to select cells B5 through H7.

2 From the MATLAB group, select Send data to MATLAB.

3 Type YAGenes for the variable name, and then click OK.

The variable YAGenes is added to the MATLAB Workspace as a 3-by-7

matrix.
4 From the MATLAB group, select Run MATLAB command.
5 Type plot(YAGenes') for the command, and then click OK.

A Figure window displays a plot of the data.

1-29

Getting Started

Note Make sure you use the ' (transpose) symbol when plotting the data
in this step. You need to transpose the data in YAGenes so that it plots as
three genes over seven time intervals.
6 Select cell J20, and then click from the MATLAB group, select Get

MATLAB figure.
The figure is added to the spreadsheet.

1-30

Get Information from Web Database

Get Information from Web Database

In this section...
What Are get Functions? on page 1-31
Creating the getpubmed Function on page 1-32

What Are get Functions?

Bioinformatics Toolbox includes several get functions that retrieve
information from various Web databases. Additionally, with some basic
MATLAB programming skills, you can create your own get function to
retrieve information from a specific Web database.
The following procedure illustrates how to create a function to retrieve
information from the NCBI PubMed database and read the information into
a MATLAB structure. The NCBI PubMed database contains biomedical
literature citations and abstracts.

1-31

Getting Started

Creating the getpubmed Function

The following procedure shows you how to create a function named getpubmed
using the MATLAB Editor. This function will retrieve citation and abstract
information from PubMed literature searches and write the data to a
MATLAB structure.
Specifically, this function will take one or more search terms, submit them
to the PubMed database for a search, then return a MATLAB structure or
structure array, with each structure containing information for an article
found by the search. The returned information will include a PubMed
identifier, publication date, title, abstract, authors, and citation.
The function will also include property name/property value pairs that let
the user of the function limit the search by publication date and limit the
number of records returned.
1 From MATLAB, open the MATLAB Editor by selecting File > New >

Function.
2 Define the getpubmed function, its input arguments, and return values by

typing:
function pmstruct = getpubmed(searchterm,varargin)
% GETPUBMED Search PubMed database & write results to MATLAB structure
3 Add code to do some basic error checking for the required input SEARCHTERM.

% Error checking for required input SEARCHTERM

if(nargin<1)
error('GETPUBMED:NotEnoughInputArguments',...
'SEARCHTERM is missing.');
end
4 Create variables for the two property name/property value pairs, and set

their default values.

% Set default settings for property name/value pairs,
% 'NUMBEROFRECORDS' and 'DATEOFPUBLICATION'
maxnum = 50; % NUMBEROFRECORDS default is 50
pubdate = ''; % DATEOFPUBLICATION default is an empty string

1-32

Get Information from Web Database

5 Add code to parse the two property name/property value pairs if provided

as input.
% Parsing the property name/value pairs
num_argin = numel(varargin);
for n = 1:2:num_argin
arg = varargin{n};
switch lower(arg)
% If NUMBEROFRECORDS is passed, set MAXNUM
case 'numberofrecords'
maxnum = varargin{n+1};
% If DATEOFPUBLICATION is passed, set PUBDATE
case 'dateofpublication'
pubdate = varargin{n+1};
end
end
6 You access the PubMed database through a search URL, which submits

a search term and options, and then returns the search results in a
specified format. This search URL is comprised of a base URL and defined
parameters. Create a variable containing the base URL of the PubMed
database on the NCBI Web site.
% Create base URL for PubMed db site
baseSearchURL = 'http://www.ncbi.nlm.nih.gov/sites/entrez?cmd=search';
7 Create variables to contain five defined parameters that the getpubmed

function will use, namely, db (database), term (search term), report (report
type, such as MEDLINE), format (format type, such as text), and dispmax
(maximum number of records to display).
% Set db parameter to pubmed
dbOpt = '&db=pubmed';
% Set term parameter to SEARCHTERM and PUBDATE
% (Default PUBDATE is '')
termOpt = ['&term=',searchterm,'+AND+',pubdate];

1-33

Getting Started

% Set report parameter to medline

reportOpt = '&report=medline';
% Set format parameter to text
formatOpt = '&format=text';
% Set dispmax to MAXNUM
% (Default MAXNUM is 50)
maxOpt = ['&dispmax=',num2str(maxnum)];
8 Create a variable containing the search URL from the variables created

in the previous steps.

% Create search URL
searchURL = [baseSearchURL,dbOpt,termOpt,reportOpt,formatOpt,maxOpt];
9 Use the urlread function to submit the search URL, retrieve the search

results, and return the results (as text in the MEDLINE report type) in
medlineText, a character array.
medlineText = urlread(searchURL);
10 Use the MATLAB regexp function and regular expressions to parse and

extract the information in medlineText into hits, a cell array, where each
cell contains the MEDLINE-formatted text for one article. The first input
is the character array to search, the second input is a search expression,
which tells the regexp function to find all records that start with PMID-,
while the third input, 'match', tells the regexp function to return the
actual records, rather than the positions of the records.
hits = regexp(medlineText,'PMID-.*?(?=PMID|</pre>$)','match');
11 Instantiate the pmstruct structure returned by getpubmed to contain six

fields.
pmstruct = struct('PubMedID','','PublicationDate','','Title','',...
'Abstract','','Authors','','Citation','');
12 Use the MATLAB regexp function and regular expressions to loop through

each article in hits and extract the PubMed ID, publication date, title,

1-34

Get Information from Web Database

abstract, authors, and citation. Place this information in the pmstruct

structure array.
for n = 1:numel(hits)
pmstruct(n).PubMedID = regexp(hits{n},'(?<=PMID- ).*?(?=\n)','match', 'once');
pmstruct(n).PublicationDate = regexp(hits{n},'(?<=DP
pmstruct(n).Title = regexp(hits{n},'(?<=TI

- ).*?(?=\n)','match', 'once');

- ).*?(?=PG

pmstruct(n).Abstract = regexp(hits{n},'(?<=AB
pmstruct(n).Authors = regexp(hits{n},'(?<=AU
pmstruct(n).Citation = regexp(hits{n},'(?<=SO

-|AB

- ).*?(?=AD

-)','match', 'once');

-)','match', 'once');

- ).*?(?=\n)','match');
- ).*?(?=\n)','match', 'once');

end

13 Select File > Save As.

When you are done, your file should look similar to the getpubmed.m
file included with the Bioinformatics Toolbox software. The sample
getpubmed.m file, including help, is located at:
matlabroot\toolbox\bioinfo\biodemos\getpubmed.m

Note The notation matlabroot is the MATLAB root directory, which is

the directory where the MATLAB software is installed on your system.

1-35

1-36

Getting Started

2
High-Throughput Sequence
Analysis
Work with Large Multi-Entry Text Files on page 2-2
Manage Short-Read Sequence Data in Objects on page 2-8
Store and Manage Feature Annotations in Objects on page 2-23
Visualize and Investigate Short-Read Alignments on page 2-31
Identifying Differentially Expressed Genes from RNA-Seq Data on page
2-44
Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data
on page 2-61
Exploring Genome-wide Differences in DNA Methylation Profiles on page
2-78

High-Throughput Sequence Analysis

Work with Large Multi-Entry Text Files

In this section...
Overview on page 2-2
What Files Can You Access? on page 2-2
Before You Begin on page 2-3
Create a BioIndexedFile Object to Access Your Source File on page 2-4
Determine the Number of Entries Indexed By a BioIndexedFile Object
on page 2-5
Retrieve Entries from Your Source File on page 2-5
Read Entries from Your Source File on page 2-6

Overview
Many biological experiments produce huge data files that are difficult to
access due to their size, which can cause memory issues when reading the
file into the MATLAB Workspace. You can construct a BioIndexedFile
object to access the contents of a large text file containing nonuniform size
entries, such as sequences, annotations, and cross-references to data sets.
The BioIndexedFile object lets you quickly and efficiently access this data
without loading the source file into memory.
You can use the BioIndexedFile object to access individual entries or a
subset of entries when the source file is too big to fit into memory. You can
access entries using indices or keys. You can read and parse one or more
entries using provided interpreters or a custom interpreter function.
Use the BioIndexedFile object in conjunction with your large source file to:
Access a subset of the entries for validation or further analysis.
Parse entries using a custom interpreter function.

What Files Can You Access?

You can use the BioIndexedFile object to access large text files.

2-2

Work with Large Multi-Entry Text Files

Your source file can have these application-specific formats:

FASTA
FASTQ
SAM
Your source file can also have these general formats:
Table Tab-delimited table with multiple columns. Keys can be in any
column. Rows with the same key are considered separate entries.
Multi-row Table Tab-delimited table with multiple columns. Keys
can be in any column. Contiguous rows with the same key are considered
a single entry. Noncontiguous rows with the same key are considered
separate entries.
Flat Flat file with concatenated entries separated by a character string,
typically //. Within an entry, the key is separated from the rest of the
entry by a white space.

Before You Begin

Before constructing a BioIndexedFile object, locate your source file on your
hard drive or a local network.
When you construct a BioIndexedFile object from your source file for the
first time, you also create an auxiliary index file, which by default is saved
to the same location as your source file. However, if your source file is in a
read-only location, you can specify a different location to save the index file.
Tip If you construct a BioIndexedFile object from your source file on
subsequent occasions, it takes advantage of the existing index file, which
saves time. However, the index file must be in the same location or a location
specified by the subsequent construction syntax.

2-3

High-Throughput Sequence Analysis

Tip If insufficient memory is not an issue when accessing your source file,
you may want to try an appropriate read function, such as genbankread, for
importing data from GenBank files. .
Additionally, several read functions such as fastaread, fastqread, samread,
and sffread include a Blockread property, which lets you read a subset of
entries from a file, thus saving memory.

Create a BioIndexedFile Object to Access Your Source

File
To construct a BioIndexedFile object from a multi-row table file:
1 Create a variable containing the full absolute path of your source file. For

your source file, use the yeastgenes.sgd file, which is included with the
Bioinformatics Toolbox software.
sourcefile = which('yeastgenes.sgd');
2 Use the BioIndexedFile constructor function to construct a

BioIndexedFile object from the yeastgenes.sgd source file, which is a

multi-row table file. Save the index file in the Current Folder. Indicate that
the source file keys are in column 3. Also, indicate that the header lines in
the source file are prefaced with !, so the constructor ignores them.
gene2goObj = BioIndexedFile('mrtab', sourcefile, '.', ...
'KeyColumn', 3, 'HeaderPrefix','!')

The BioIndexedFile constructor function constructs gene2goObj, a

BioIndexedFile object, and also creates an index file with the same name
as the source file, but with an IDX extension. It stores this index file in the
Current Folder because we specified this location. However, the default
location for the index file is the same location as the source file.
Caution Do not modify the index file. If you modify it, you can get invalid
results. Also, the constructor function cannot use a modified index file to
construct future objects from the associated source file.

2-4

Work with Large Multi-Entry Text Files

Determine the Number of Entries Indexed By a

BioIndexedFile Object
To determine the number of entries indexed by a BioIndexedFile object, use
the NumEntries property of the BioIndexedFile object. For example, for
the gene2goObj object:
gene2goObj.NumEntries
ans =
6476

Note For a list and description of all properties of a BioIndexedFile object,

see BioIndexedFile class.

Retrieve Entries from Your Source File

Retrieve entries from your source file using either:
The index of the entry
The entry key

Retrieve Entries Using Indices

Use the getEntryByIndex method to retrieve a subset of entries from your
source file that correspond to specified indices. For example, retrieve the first
12 entries from the yeastgenes.sgd source file:
subset_entries = getEntryByIndex(gene2goObj, [1:12]);

Retrieve Entries Using Keys

Use the getEntryByKey method to retrieve a subset of entries from your
source file that are associated with specified keys. For example, retrieve all
entries with keys of AAC1 and AAD10 from the yeastgenes.sgd source file:
subset_entries = getEntryByKey(gene2goObj, {'AAC1' 'AAD10'});

2-5

High-Throughput Sequence Analysis

The output subset_entries is a single string of concatenated entries.

Because the keys in the yeastgenes.sgd source file are not unique, this
method returns all entries that have a key of AAC1 or AAD10.

Read Entries from Your Source File

The BioIndexedFile object includes a read method, which you can use to
read and parse a subset of entries from your source file. The read method
parses the entries using an interpreter function specified by the Interpreter
property of the BioIndexedFile object.

Set the Interpreter Property

Before using the read method, make sure the Interpreter property of the
BioIndexedFile object is set appropriately.
If you constructed a
BioIndexedFile object from
...

The Interpreter property ...

A source file with an

application-specific format
(FASTA, FASTQ, or SAM)

By default is a handle to a function

appropriate for that file type and
typically does not require you to change
it.

A source file with a table,

multi-row table, or flat format

By default is [], which means the

interpreter is an anonymous function
in which the output is equivalent to the
input. You can change this to a handle
to a function that accepts a single string
of one or more concatenated entries
and returns a structure or an array of
structures containing the interpreted
data.

There are two ways to set the Interpreter property of the BioIndexedFile
object:
When constructing the BioIndexedFile object, use the Interpreter
property name/property value pair

2-6

Work with Large Multi-Entry Text Files

After constructing the BioIndexedFile object, set the Interpreter

property
Note For more information on setting the Interpreter property of a
BioIndexedFile object, see BioIndexedFile class.

Read a Subset of Entries

The read method reads and parses a subset of entries that you specify using
either entry indices or keys.

Example
To quickly find all the gene ontology (GO) terms associated with a particular
gene because the entry keys are gene names:
1 Set the Interpreter property of the gene2goObj BioIndexedFile object

to a handle to a function that reads entries and returns only the column
containing the GO term. In this case the interpreter is a handle to an
anonymous function that accepts strings and extracts strings that start
with the characters GO.
gene2goObj.Interpreter = @(x) regexp(x,'GO:\d+','match')
2 Read only the entries that have a key of YAT2, and return their GO terms.

GO_YAT2_entries = read(gene2goObj, 'YAT2')

GO_YAT2_entries =
'GO:0004092' 'GO:0005737' 'GO:0006066' 'GO:0006066' 'GO:0009437'

2-7

High-Throughput Sequence Analysis

Manage Short-Read Sequence Data in Objects

In this section...
Overview on page 2-8
Represent Sequence and Quality Data in a BioRead Object on page 2-9
Represent Sequence, Quality, and Alignment/Mapping Data in a BioMap
Object on page 2-11
Retrieve Information from a BioRead or BioMap Object on page 2-15
Set Information in a BioRead or BioMap Object on page 2-17
Determine Coverage of a Reference Sequence on page 2-18
Construct Sequence Alignments to a Reference Sequence on page 2-20
Filter Read Sequences Using SAM Flags on page 2-21

Overview
High-throughput sequencing instruments produce large amounts of short-read
sequence data that can be challenging to store and manage. Using objects to
contain this data lets you easily access, manipulate, and filter the data.
Bioinformatics Toolbox includes two objects for working with short-read
sequence data.
Object

Contains This Information

Construct from One of

These

BioRead

Sequence headers

FASTQ file

Read sequences

SAM file

Sequence qualities (base

calling)

FASTQ structure (created

using the fastqread
function)
SAM structure (created
using the samread function)
Cell arrays containing
header, sequence, and

2-8

Manage Short-Read Sequence Data in Objects

Object

Contains This Information

Construct from One of

These
quality information
(created using the
fastqread function)

BioMap

Sequence headers

SAM file

Read sequences

BAM file

Sequence qualities (base

calling)

SAM structure (created

using the samread function)

Sequence alignment and

mapping information
(relative to a single
reference sequence),
including mapping quality

BAM structure (created

using the bamread function)
Cell arrays containing
header, sequence, quality,
and mapping/alignment
information (created using
the samread or bamread
function)

Represent Sequence and Quality Data in a BioRead

Object
Prerequisites
A BioRead object represents a collection of short-read sequences. Each
element in the object is associated with a sequence, sequence header, and
sequence quality information.
Construct a BioRead object in one of two ways:
Indexed The data remains in the source file. Constructing the object
and accessing its contents is memory efficient. However, you cannot modify
object properties, other than the Name property. This is the default method
if you construct a BioRead object from a FASTQ- or SAM-formatted file.
In Memory The data is read into memory. Constructing the object
and accessing its contents is limited by the amount of available memory.

2-9

High-Throughput Sequence Analysis

However, you can modify object properties. When you construct a BioRead
object from a FASTQ structure or cell arrays, the data is read into memory.
When you construct a BioRead object from a FASTQ- or SAM-formatted file,
use the InMemory name-value pair argument to read the data into memory.

Construct a BioRead Object from a FASTQ- or SAM-Formatted

File
Note This example constructs a BioRead object from a FASTQ-formatted file.
Use similar steps to construct a BioRead object from a SAM-formatted file.
Use the BioRead constructor function to construct a BioRead object from a
FASTQ-formatted file and set the Name property:
BRObj1 = BioRead('SRR005164_1_50.fastq', 'Name', 'MyObject')
BRObj1 =
BioRead with properties:
Quality:
Sequence:
Header:
NSeqs:
Name:

[50x1 File indexed property]

[50x1 File indexed property]
[50x1 File indexed property]
50
'MyObject'

The constructor function construct a BioRead object and, if an index file does
not already exist, it also creates an index file with the same file name, but
with an .IDX extension. This index file, by default, is stored in the same
location as the source file.

2-10

Manage Short-Read Sequence Data in Objects

Caution

Your source file and index file must always be in sync.

After constructing a BioRead object, do not modify the index file, or you
can get invalid results when using the existing object or constructing new
objects.
If you modify the source file, delete the index file, so the object constructor
creates a new index file when constructing new objects.

Note Because you constructed this BioRead object from a source file, you
cannot modify the properties (except for Name) of the BioRead object.

Represent Sequence, Quality, and

Alignment/Mapping Data in a BioMap Object
Prerequisites
A BioMap object represents a collection of short-read sequences that map
against a single reference sequence. Each element in the object is associated
with a read sequence, sequence header, sequence quality information, and
alignment/mapping information.
When constructing a BioMap object from a BAM file, the maximum size of the
file is limited by your operating system and available memory.
Construct a BioMap object in one of two ways:
Indexed The data remains in the source file. Constructing the object
and accessing its contents is memory efficient. However, you cannot modify
object properties, other than the Name property. This is the default method
if you construct a BioMap object from a SAM- or BAM-formatted file.
In Memory The data is read into memory. Constructing the object
and accessing its contents is limited by the amount of available memory.
However, you can modify object properties. When you construct a BioMap
object from a structure, the data stays in memory. When you construct

2-11

High-Throughput Sequence Analysis

a BioMap object from a SAM- or BAM-formatted file, use the InMemory

name-value pair argument to read the data into memory.

Construct a BioMap Object from a SAM- or BAM-Formatted File

Note This example constructs a BioMap object from a SAM-formatted file.
Use similar steps to construct a BioMap object from a BAM-formatted file.
1 If you do not know the number and names of the reference sequences in

your source file, determine them using the saminfo or baminfo function
and the ScanDictionary name-value pair argument.
samstruct = saminfo('ex2.sam', 'ScanDictionary', true);
samstruct.ScannedDictionary
ans =
'seq1'
'seq2'

Tip The previous syntax scans the entire SAM file, which is time
consuming. If you are confident that the Header information of the SAM
file is correct, omit the ScanDictionary name-value pair argument, and
inspect the SequenceDictionary field instead.
2 Use the BioMap constructor function to construct a BioMap object from

the SAM file and set the Name property. Because the SAM-formatted file
in this example, ex2.sam, contains multiple reference sequences, use the
SelectRef name-value pair argument to specify one reference sequence,
seq1:
BMObj2 = BioMap('ex2.sam', 'SelectRef', 'seq1', 'Name', 'MyObject')

BMObj2 =
BioMap with properties:

2-12

Manage Short-Read Sequence Data in Objects

SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

'seq1'
[1501x1 File
[1501x1 File
[1501x1 File
[1501x1 File
[1501x1 File
[1501x1 File
[1501x1 File
[1501x1 File
[1501x1 File
1501
'MyObject'

indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed

property]
property]
property]
property]
property]
property]
property]
property]
property]

The constructor function constructs a BioMap object and, if index files do not
already exist, it also creates one or two index files:
If constructing from a SAM-formatted file, it creates one index file that has
the same file name as the source file, but with an .IDX extension. This
index file, by default, is stored in the same location as the source file.
If constructing from a BAM-formatted file, it creates two index files that
have the same file name as the source file, but one with a .BAI extension
and one with a .LINEARINDEX extension. These index files, by default,
are stored in the same location as the source file.

Caution

Your source file and index files must always be in sync.

After constructing a BioMap object, do not modify the index files, or you
can get invalid results when using the existing object or constructing new
objects.
If you modify the source file, delete the index files, so the object constructor
creates new index files when constructing new objects.

2-13

High-Throughput Sequence Analysis

Note Because you constructed this BioMap object from a source file, you
cannot modify the properties (except for Name and Reference) of the BioMap
object.

Construct a BioMap Object from a SAM or BAM Structure

Note This example constructs a BioMap object from a SAM structure using
samread. Use similar steps to construct a BioMap object from a BAM structure
using bamread.
1 Use the samread function to create a SAM structure from a SAM-formatted

file:
SAMStruct = samread('ex2.sam');
2 To construct a valid BioMap object from a SAM-formatted file, the file must

contain only one reference sequence. Determine the number and names
of the reference sequences in your SAM-formatted file using the unique
function to find unique names in the ReferenceName field of the structure:
unique({SAMStruct.ReferenceName})
ans =
'seq1'

'seq2'

3 Use the BioMap constructor function to construct a BioMap object from a

SAM structure. Because the SAM structure contains multiple reference

sequences, use the SelectRef name-value pair argument to specify one
reference sequence, seq1:
BMObj1 = BioMap(SAMStruct, 'SelectRef', 'seq1')
BMObj1 =
BioMap with properties:

2-14

Manage Short-Read Sequence Data in Objects

SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

{'seq1'}
{1501x1 cell}
{1501x1 cell}
[1501x1 uint32]
[1501x1 uint8]
[1501x1 uint16]
[1501x1 uint32]
{1501x1 cell}
{1501x1 cell}
{1501x1 cell}
1501
''

Retrieve Information from a BioRead or BioMap

Object
You can retrieve all or a subset of information from a BioRead or BioMap
object.

Retrieve a Property from a BioRead or BioMap Object

You can retrieve a specific property from elements in a BioRead or BioMap
object.
For example, to retrieve all headers from a BioRead object, use the Header
property as follows:
allHeaders = BRObj1.Header;

This syntax returns a cell array containing the headers for all elements in the
BioRead object.
Similarly, to retrieve all start positions of aligned read sequences from a
BioMap object, use the Start property of the object:
allStarts = BMObj1.Start;

This syntax returns a vector containing the start positions of aligned read
sequences with respect to the position numbers in the reference sequence in
a BioMap object.

2-15

High-Throughput Sequence Analysis

Retrieve Multiple Properties from a BioRead or BioMap Object

You can retrieve multiple properties from a BioRead or BioMap object in a
single command using the get method. For example, to retrieve both start
positions and headers information of a BioMap object, use the get method as
follows:
multiProp = get(BMObj1, {'Start', 'Header'});

This syntax returns a cell array containing all start positions and headers
information of a BioMap object.
Note Property names are case sensitive.
For a list and description of all properties of a BioRead object, see BioRead
class. For a list and description of all properties of a BioMap object, see BioMap
class.

Retrieve a Subset of Information from a BioRead or BioMap

Object
Use specialized get methods with a numeric vector, logical vector, or cell
array of headers to retrieve a subset of information from an object. For
example, to retrieve the first 10 elements from a BioRead object, use the
getSubset method:
newBRObj = getSubset(BRObj1, [1:10]);

This syntax returns a new BioRead object containing the first 10 elements in
the original BioRead object.
For example, to retrieve the first 12 positions of sequences with headers
SRR005164.1, SRR005164.7, and SRR005164.16, use the getSubsequence
method:
subSeqs = getSubsequence(BRObj1, ...
{'SRR005164.1', 'SRR005164.7', 'SRR005164.16'}, [1:12]')
subSeqs =

2-16

Manage Short-Read Sequence Data in Objects

'TGGCTTTAAAGC'
'CCCGAAAGCTAG'
'AATTTTGCGGCT'

For example, to retrieve information about the third element in a BioMap

object, use the getInfo method:
Info_3 = getInfo(BMObj1, 3);

This syntax returns a tab-delimited string containing this information for

the third element:
Sequence header
SAM flags for the sequence
Start position of the aligned read sequence with respect to the reference
sequence
Mapping quality score for the sequence
Signature (CIGAR-formatted string) for the sequence
Sequence
Quality scores for sequence positions
Note Method names are case sensitive.
For a complete list and description of methods of a BioRead object, see
BioRead class. For a complete list and description of methods of a BioMap
object, see BioMap class.

Set Information in a BioRead or BioMap Object

Prerequisites
To modify properties (other than Name and Reference) of a BioRead or BioMap
object, the data must be in memory, and not indexed. To ensure the data is
in memory, do one of the following:

2-17

High-Throughput Sequence Analysis

Construct the object from a structure as described in Construct a BioMap

Object from a SAM or BAM Structure on page 2-14.
Construct the object from a source file using the InMemory name-value
pair argument.

Provide Custom Headers for Sequences

First, create an object with the data in memory:
BRObj1 = BioRead('SRR005164_1_50.fastq','InMemory',true);

To provide custom headers for sequences of interest (in this case sequences 1
to 5), do the following:
BRObj1.Header(1:5) = {'H1', 'H2', 'H3', 'H4', 'H5'};

Alternatively, you can use the setHeader method:

BRObj1 = setHeader(BRObj1, {'H1', 'H2', 'H3', 'H4', 'H5'}, [1:5]);

Several other specialized set methods let you set the properties of a subset of
elements in a BioRead or BioMap object.
Note Method names are case sensitive.
For a complete list and description of methods of a BioRead object, see
BioRead class. For a complete list and description of methods of a BioMap
object, see BioMap class.

Determine Coverage of a Reference Sequence

When working with a BioMap object, you can determine the number of read
sequences that:
Align within a specific region of the reference sequence
Align to each position within a specific region of the reference sequence

2-18

Manage Short-Read Sequence Data in Objects

For example, you can compute the number, indices, and start positions of
the read sequences that align within the first 25 positions of the reference
sequence. To do so, use the getCounts, getIndex, and getStart methods:
Cov = getCounts(BMObj1, 1, 25)
Cov =
12
Indices = getIndex(BMObj1, 1, 25)
Indices =
1
2
3
4
5
6
7
8
9
10
11
12
startPos = getStart(BMObj1, Indices)
startPos =
1
3
5
6
9
13
13
15
18
22

2-19

High-Throughput Sequence Analysis

22
24

The first two syntaxes return the number and indices of the read sequences
that align within the specified region of the reference sequence. The last
syntax returns a vector containing the start position of each aligned read
sequence, corresponding to the position numbers of the reference sequence.
For example, you can also compute the number of the read sequences that
align to each of the first 10 positions of the reference sequence. For this
computation, use the getBaseCoverage method:
Cov = getBaseCoverage(BMObj1, 1, 10)
Cov =
1

Construct Sequence Alignments to a Reference

Sequence
It is useful to construct and view the alignment of the read sequences that
align to a specific region of the reference sequence. It is also helpful to know
which read sequences align to this region in a BioMap object.
For example, to retrieve the alignment of read sequences to the first 12
positions of the reference sequence in a BioMap object, use the getAlignment
method:
[Alignment_1_12, Indices] = getAlignment(BMObj2, 1, 12)
Alignment_1_12 =
CACTAGTGGCTC
CTAGTGGCTC
AGTGGCTC
GTGGCTC
GCTC

Indices =

2-20

Manage Short-Read Sequence Data in Objects

1
2
3
4
5

Return the headers of the read sequences that align to a specific region of
the reference sequence:
alignedHeaders = getHeader(BMObj2, Indices)
alignedHeaders =
'B7_591:4:96:693:509'
'EAS54_65:7:152:368:113'
'EAS51_64:8:5:734:57'
'B7_591:1:289:587:906'
'EAS56_59:8:38:671:758'

Filter Read Sequences Using SAM Flags

SAM- and BAM-formatted files include the status of 11 binary flags for each
read sequence. These flags describe different sequencing and alignment
aspects of a read sequence. For more information on the flags, see theSAM
Format Specification. The filterByFlag method lets you filter the read
sequences in a BioMap object by using these flags.

Filter Unmapped Read Sequences

1 Construct a BioMap object from a SAM-formatted file.

BMObj2 = BioMap('ex1.sam');
2 Use the filterByFlag method to create a logical vector indicating the read

sequences in a BioMap object that are mapped.

LogicalVec_mapped = filterByFlag(BMObj2, 'unmappedQuery', false);
3 Use this logical vector and the getSubset method to create a new BioMap

object containing only the mapped read sequences.

2-21

High-Throughput Sequence Analysis

filteredBMObj_1 = getSubset(BMObj2, LogicalVec_mapped);

Filter Read Sequences That Are Not Mapped in a Pair

1 Construct a BioMap object from a SAM-formatted file.

BMObj2 = BioMap('ex1.sam');
2 Use the filterByFlag method to create a logical vector indicating the read

sequences in a BioMap object that are mapped in a proper pair, that is, both
the read sequence and its mate are mapped to the reference sequence.
LogicalVec_paired = filterByFlag(BMObj2, 'pairedInMap', true);
3 Use this logical vector and the getSubset method to create a new BioMap

object containing only the read sequences that are mapped in a proper pair.
filteredBMObj_2 = getSubset(BMObj2, LogicalVec_paired);

2-22

Store and Manage Feature Annotations in Objects

Store and Manage Feature Annotations in Objects

In this section...
Represent Feature Annotations in a GFFAnnotation or GTFAnnotation
Object on page 2-23
Construct an Annotation Object on page 2-23
Retrieve General Information from an Annotation Object on page 2-24
Access Data in an Annotation Object on page 2-25
Use Feature Annotations with Short-Read Sequence Data on page 2-26

Represent Feature Annotations in a GFFAnnotation

or GTFAnnotation Object
The GFFAnnotation and GTFAnnotation objects represent a collection of
feature annotations for one or more reference sequences. You construct
these objects from GFF (General Feature Format) and GTF (Gene Transfer
Format) files. Each element in the object represents a single annotation.
The properties and methods associated with the objects let you investigate
and filter the data based on reference sequence, a feature (such as CDS or
exon), or a specific gene or transcript.

Construct an Annotation Object

Use the GFFAnnotation constructor function to construct a GFFAnnotation
object from either a GFF- or GTF-formatted file:
GFFAnnotObj = GFFAnnotation('tair8_1.gff')
GFFAnnotObj =
GFFAnnotation with properties:
FieldNames: {1x9 cell}
NumEntries: 3331

Use the GTFAnnotation constructor function to construct a GTFAnnotation

object from a GTF-formatted file:

2-23

High-Throughput Sequence Analysis

GTFAnnotObj = GTFAnnotation('hum37_2_1M.gtf')
GTFAnnotObj =
GTFAnnotation with properties:
FieldNames: {1x11 cell}
NumEntries: 308

Retrieve General Information from an Annotation

Object
Determine the field names and the number of entries in an annotation object
by accessing the FieldNames and NumEntries properties. For example, to see
the field names for each annotation object constructed in the previous section,
query the FieldNames property:
GFFAnnotObj.FieldNames
ans =
Columns 1 through 6
'Reference'

'Start'

'Stop'

'Feature'

'Source'

'Feature'

'Gene'

'Score'

Columns 7 through 9
'Strand'

'Frame'

'Attributes'

GTFAnnotObj.FieldNames
ans =
Columns 1 through 6
'Reference'

'Start'

'Stop'

Columns 7 through 11
'Source'

2-24

'Score'

'Strand'

'Frame'

'Attributes'

'Transcript'

Store and Manage Feature Annotations in Objects

Determine the range of the reference sequences that are covered by feature
annotations by using the getRange method with the annotation object
constructed in the previous section:
range = getRange(GFFAnnotObj)
range =
3631

498516

Access Data in an Annotation Object

Create a Structure of the Annotation Data
Creating a structure of the annotation data lets you access the field values.
Use the getData method to create a structure containing a subset of the data
in a GFFAnnotation object constructed in the previous section.
% Extract annotations for positions 1 through 10000 of the
% reference sequence
AnnotStruct = getData(GFFAnnotObj,1,10000)
AnnotStruct =
60x1 struct array with fields:
Reference
Start
Stop
Feature
Source
Score
Strand
Frame
Attributes

Access Field Values in the Structure

Use dot indexing to access all or specific field values in a structure.
For example, extract the start positions for all annotations:

2-25

High-Throughput Sequence Analysis

Starts = AnnotStruct.Start;

Extract the start positions for annotations 12 through 17. Notice that you
must use square brackets when indexing a range of positions:
Starts_12_17 = [AnnotStruct(12:17).Start]
Starts_12_17 =
4706

5174

5174

5439

5439

5631

Extract the start position and the feature for the 12th annotation:
Start_12 = AnnotStruct(12).Start
Start_12 =
4706
Feature_12 = AnnotStruct(12).Feature
Feature_12 =
CDS

Use Feature Annotations with Short-Read Sequence

Data
Investigate the results of short-read sequence experiments by using
GFFAnnotation and GTFAnnotation objects with BioMap objects. For example,
you can:
Determine counts of short-read sequences aligned to regions of a reference
sequence associated with specific annotations, such as in RNA-Seq
workflows.
Find annotations within a specific range of a peak of interest in a reference
sequence, such as in ChIP-Seq workflows.

Determine Annotations of Interest

1 Construct a GTFAnnotation object from a GTF- formatted file:

2-26

Store and Manage Feature Annotations in Objects

GTFAnnotObj = GTFAnnotation('hum37_2_1M.gtf');
2 Use the getReferenceNames method to return the names for the reference

sequences for the annotation object:

refNames = getReferenceNames(GTFAnnotObj)
refNames =
'chr2'
3 Use the getFeatureNames method to retrieve the feature names from the

annotation object:
featureNames = getFeatureNames(GTFAnnotObj)
featureNames =
'CDS'
'exon'
'start_codon'
'stop_codon'
4 Use the getGeneNames method to retrieve a list of the unique gene names

from the annotation object:

geneNames = getGeneNames(GTFAnnotObj)
geneNames =
'uc002qvu.2'
'uc002qvv.2'
'uc002qvw.2'
'uc002qvx.2'
'uc002qvy.2'
'uc002qvz.2'
'uc002qwa.2'
'uc002qwb.2'
'uc002qwc.1'
'uc002qwd.2'
'uc002qwe.3'

2-27

High-Throughput Sequence Analysis

'uc002qwf.2'
'uc002qwg.2'
'uc002qwh.2'
'uc002qwi.3'
'uc002qwk.2'
'uc002qwl.2'
'uc002qwm.1'
'uc002qwn.1'
'uc002qwo.1'
'uc002qwp.2'
'uc002qwq.2'
'uc010ewe.2'
'uc010ewf.1'
'uc010ewg.2'
'uc010ewh.1'
'uc010ewi.2'
'uc010yim.1'

The previous steps gave us a list of available reference sequences, features,

and genes associated with the available annotations. Use this information
to determine annotations of interest. For instance, you might be interested
only in annotations that are exons associated with the uc002qvv.2 gene on
chromosome 2.

Filter Annotations
Use the getData method to filter the annotations and create a structure
containing only the annotations of interest, which are annotations that are
exons associated with the uc002qvv.2 gene on chromosome 2.
AnnotStruct = getData(GTFAnnotObj,'Reference','chr2',...
'Feature','exon','Gene','uc002qvv.2')
AnnotStruct =
12x1 struct array with fields:
Reference
Start
Stop
Feature

2-28

Store and Manage Feature Annotations in Objects

Gene
Transcript
Source
Score
Strand
Frame
Attributes

The return structure contains 12 elements, indicating there are 12

annotations that meet your filter criteria.

Extract Position Ranges for Annotations of Interest

After filtering the data to include only annotations that are exons associated
with the uc002qvv.2 gene on chromosome 2, use the Start and Stop fields to
create vectors of the start and end positions for the ranges associated with the
12 annotations.
StartPos = [AnnotStruct.Start];
EndPos = [AnnotStruct.Stop];

Determine Counts of Short-Read Sequences Aligned to

Annotations
Construct a BioMap object from a BAM-formatted file containing short-read
sequence data aligned to chromosome 2.
BMObj3 = BioMap('ex3.bam');

Then use the range for the annotations of interest as input to the getCounts
method of a BioMap object. This returns the counts of short reads aligned to
the annotations of interest.
counts = getCounts(BMObj3,StartPos,EndPos,'independent', true)
counts =
1399
1
54
221
97

2-29

High-Throughput Sequence Analysis

125
0
1
0
65
9
12

2-30

Visualize and Investigate Short-Read Alignments

Visualize and Investigate Short-Read Alignments

In this section...
When to Use the NGS Browser to Visualize and Investigate Data on page
2-31
Open the NGS Browser on page 2-32
Import Data into the NGS Browser on page 2-33
Zoom and Pan to a Specific Region of the Alignment on page 2-36
View Coverage of the Reference Sequence on page 2-37
View the Pileup View of Short Reads on page 2-38
Compare Alignments of Multiple Data Sets on page 2-39
View Location, Quality Scores, and Mapping Information on page 2-40
Flag Reads on page 2-41
Evaluate and Flag Mismatches on page 2-42
View Insertions and Deletions on page 2-43
View Feature Annotations on page 2-43
Print and Export the Browser Image on page 2-43

When to Use the NGS Browser to Visualize and

Investigate Data
The NGS Browser lets you visually verify and investigate the alignment of
short-read sequences to a reference sequence, in support of analyses that
measure genetic variations and gene expression. The NGS Browser lets you:
Visualize short-read data aligned to a nucleotide reference sequence.
Compare multiple data sets aligned against a common reference sequence.
View coverage of different bases and regions of the reference sequence.
Investigate quality and other details of aligned reads.
Identify mismatches due to base calling errors or polymorphisms.
Visualize insertions and deletions.

2-31

High-Throughput Sequence Analysis

Retrieve feature annotations relative to a specific region of the reference

sequence.
Investigate regions of interest in the alignment, determined by various
analyses.
You can visualize and investigate the aligned data before, during, or after any
preprocessing (filtering, quality recalibration) or analysis steps you perform
on the aligned data.

Open the NGS Browser

To open the NGS Browser, type the following in the MATLAB Command
Window:
ngsbrowser

Alternatively, click the NGS Browser on the Apps tab.

2-32

Visualize and Investigate Short-Read Alignments

Import Data into the NGS Browser

2-33
Browser Displaying Reference Track, One Alignment Track, and One Annotation Track

High-Throughput Sequence Analysis

Import a Reference Sequence

You can import a single reference sequence into the NGS Browser. The
reference sequence must be in a FASTA file.
1 Select File > Add Data from File.
2 In the Open dialog box, select a FASTA file, and then click Open.

Tip You can use the getgenbank function with the ToFile and SequenceOnly
name-value pair arguments to retrieve a reference sequence from the
GenBank database and save it to a FASTA-formatted file.

Import Short-Read Alignment Data

You can import multiple data sets of short-read alignment data. The
alignment data must be in either of the following:
BioMap object
Tip Construct a BioMap object from a SAM- or BAM-formatted file to
investigate, subset, and filter the data before importing it into the NGS
Browser.
SAM- or BAM-formatted file
Note Your SAM- or BAM-formatted file must:

2-34

Have reads ordered by start position in the reference sequence.

Have an IDX index file (for a SAM-formatted file) or BAI and
LINEARINDEX index files (for a BAM-formatted file) stored in the same
location as your source file. Otherwise, the source file must be stored
in a location to which you have write access, because MATLAB needs
to create and store index files in this location.

Visualize and Investigate Short-Read Alignments

Tip Try using SAMtools to check if the reads in your SAM- or

BAM-formatted file are ordered by position in the reference sequence, and
also to reorder them, if needed.

Tip If you do not have index files (IDX or BAI and LINEARINDEX) stored
in the same location as your source file, and your source file is stored in a
location to which you do not have write access, you cannot import data from
the source file directly into the browser. Instead, construct a BioMap object
from the source file using the IndexDir name-value pair argument, and
then import the BioMap object into the browser.
To import short-read alignment data:
1 Select File > Add Data from File or File > Import Alignment Data

from MATLAB Workspace.

2 Select a SAM-formatted file, BAM-formatted file, or BioMap object.
3 If you select a file containing multiple reference sequences, in the Select

Reference dialog box, select a reference or scan the file for available
references and their mapped reads counts. Click OK.
4 Repeat the previous steps to import additional data sets.

Import Feature Annotations

You can import multiple sets of feature annotations from GFF- or
GTF-formatted files that contain data for a single reference sequence.
1 Select File > Add Data from File.
2 In the Open dialog box, select a GFF- or GTF-formatted file, and then

click Open.
3 Repeat the previous steps to import additional annotations.

2-35

High-Throughput Sequence Analysis

Zoom and Pan to a Specific Region of the Alignment

To zoom in and out:
Use the
toolbar buttons,
or click-drag an edge of the rubberband in the Overview area.

To pan across the alignment:

toolbar buttons,
Use the
or click-drag the rubberband in the Overview area.

Tip Use the left and right arrow keys to pan in one base pair (bp) increments.

2-36

Visualize and Investigate Short-Read Alignments

View Coverage of the Reference Sequence

At the top of each alignment track, the coverage view displays the coverage
of each base in the reference sequence. The vertical ruler on the left edge of
the coverage view indicates the maximum coverage in the display range.
Hover the mouse pointer over a position in the coverage view to display the
location and counts.

Note The browser computes coverage at the base pair resolution, instead
of binning, even when zoomed out.
To change the percent coverage displayed, click anywhere in the alignment
track, and then edit the Alignment Coverage settings.

Tip Set Max to a value greater than 100, if needed, when comparing the
coverage of multiple tracks of reads.

2-37

High-Throughput Sequence Analysis

View the Pileup View of Short Reads

Each alignment track includes a pileup view of the short reads aligned to
the reference sequence.

Limit the depth of the reads displayed in the pileup view by setting the
Maximum display read depth in the Alignment Pileup settings.

Tip Limiting the depth of short reads in the pileup view does not change the
counts displayed in the coverage view.

2-38

Visualize and Investigate Short-Read Alignments

Compare Alignments of Multiple Data Sets

Compare multiple data sets, with each data set in its own track, against a
common reference sequence. Use the Track List to show/hide, order, and
delete tracks of data.

2-39

High-Throughput Sequence Analysis

View Location, Quality Scores, and Mapping

Information
Hover the mouse pointer over a position in a read to display strand direction,
location, quality, and mapping information for the base, the read, and its
paired mate.

2-40

Visualize and Investigate Short-Read Alignments

Flag Reads
Click anywhere in an alignment track to display the Alignment Pileup
settings.

Flag Reads with Low Mapping Quality

Set the Mapping quality threshold in the Alignment Pileup section to flag
low-quality reads. Reads with a mapping quality below this level appear in a
lighter shade of gray.

Flag Duplicate Reads

Select Flag duplicate reads and select an outline color.

Flag Reads with Unmapped Pairs

Select Flag reads with unmapped pair and select an outline color.

2-41

High-Throughput Sequence Analysis

Evaluate and Flag Mismatches

Mismatches display as colored blocks or letters, depending on the zoom level.

Zoomed out view of read Mismatches display as bars

Zoomed in view of read Mismatches display as letters

In addition to the base Phred quality information that displays in the tooltip,
you can visualize quality differences by using the Shade mismatch bases
by Phred quality settings.

The mismatch blocks or letters display in:

Light shade Mismatch bases with Phred scores below the minimum
Graduation of medium shades Mismatch bases with Phred scores within
the minimum to maximum range
Dark shade Mismatch bases with Phred scores above the maximum

2-42

Visualize and Investigate Short-Read Alignments

View Insertions and Deletions

The NGS Browser designates insertions with a
symbol. Hover the mouse
pointer over the insertion symbol to display information about it.

The NGS Browser designates deletions with dashes.

View Feature Annotations

After importing a feature annotation file, you can zoom and pan to view
feature annotations associated with a region of interest in the alignment.
Hover the mouse pointer over the feature annotation.

Print and Export the Browser Image

Print or export the browser image by selecting File > Print Image or File >
Export Image.

2-43

High-Throughput Sequence Analysis

Identifying Differentially Expressed Genes from RNA-Seq

Data
This example shows how to load RNA-seq data and test for differential
expression using a negative binomial model.
Introduction

RNA-seq is an emerging technology for surveying gene expression and

transcriptome content by directly sequencing the mRNA molecules in a
sample. RNA-seq can provide gene expression measurements and is regarded
as an attractive approach to analyze a transcriptome in an unbiased and
comprehensive manner.
In this example, you will use Bioinformatics Toolbox and Statistics
Toolbox functions to load publicly available transcriptional profiling
sequencing data into MATLAB, compute the digital gene expression, and
then identify differentially expressed genes in RNA-seq data from hormone
treated prostate cancer cell line samples [1].
The Prostate Cancer Data Set

In the prostate cancer study, the prostate cancer cell line LNCap was treated
with androgen/DHT. Mock-treated and androgen-stimulated LNCap cells
were sequenced using the Illumina 1G Genome Analyzer [1]. For the
mock-treated cells, there were four lanes totaling ~10 million reads. For the
DHT-treated cells, there were three lanes totaling ~7 million reads. All
replicates were technical replicates. Samples labeled s1 through s4 are from
mock-treated cells. Samples labeled s5, s6, and s8 are from DHT-treated
cells. The read sequences are stored in FASTA files. The sequence IDs break
down as follows: seq_(unique sequence id)_(number of times this sequence
was seen in this lane).
This example assumes that you have:
(1) Downloaded and uncompressed the seven FASTA files (s1.fa, s2.fa,
s3.fa, s4.fa, s5.fa, s6.fa and s8.fa) containing the raw, 35bp, unmapped
short reads from the authors Web Site.

2-44

Identifying Differentially Expressed Genes from RNA-Seq Data

(2) Produced a SAM-formatted file for each of the seven FASTA files by
mapping the short reads to the NCBI version 37 of the human genome using a
mapper such as Bowtie [2],
(3) Ordered the SAM-formatted files by reference name first, then by genomic
position.
For the published version of this example, 4,388,997 short reads were mapped
using the Bowtie aligner [2]. The aligner was instructed to report one best
valid alignment. No more than two mismatches were allowed for alignment.
Reads with more than one reportable alignment were suppressed, i.e. any
read that mapped to multiple locations was discarded. The alignment was
output to seven SAM files (s1.sam, s2.sam, s3.sam, s4.sam, s5.sam, s6.sam
and s8.sam). Because the input files were FASTA files, all quality values
were assumed to be 40 on the Phred quality scale [2]. We then used SAMtools
[3] to sort the mapped reads in the seven SAM files, one for each replicate.
Creating an Annotation Object of Target Genes

Download from Ensembl a tab-separated-value (TSV) table with all protein

encoding genes to a text file, ensemblmart_genes_hum37.txt. For this
example, we are using Ensembl release 64. Using Ensembls BioMart service,
you can select a table with the following attributes: chromosome name, gene
biotype, gene name, gene start/end, and strand direction.
Use the provided helper function ensemblmart2gff to convert the downloaded
TSV file to a GFF formatted file. Then use GFFAnnotation to load the file
into MATLAB.
GFFfilename = ensemblmart2gff('ensemblmart_genes_hum37.txt');
genes = GFFAnnotation(GFFfilename)

genes =
GFFAnnotation with properties:
FieldNames: {1x9 cell}
NumEntries: 21184

2-45

High-Throughput Sequence Analysis

Create a subset with the genes present in chromosomes only (without contigs).
The GFFAnnotation object contais 20012 annotated protein-coding genes in
the Ensembl database.
chrs = {'1','2','3','4','5','6','7','8','9','10','11','12','13','14',...
'15','16','17','18','19','20','21','22','X','Y','MT'};
genes = getSubset(genes,'reference',chrs)

genes =
GFFAnnotation with properties:
FieldNames: {1x9 cell}
NumEntries: 20012

Copy the gene information into a structure and display the first entry.
getData(genes,1)

ans =
Reference:
Start:
Stop:
Feature:
Source:
Score:
Strand:
Frame:
Attributes:

'1'
205111632
205180727
'DSTYK'
'protein_coding'
'0.0'
'-'
'.'
''

Importing Mapped Short Read Alignment Data

The size of the sorted SAM files in this data set are in the order of 250-360MB.
You can access the mapped reads in s1.sam by creating a BioMap. BioMap has

2-46

Identifying Differentially Expressed Genes from RNA-Seq Data

an interface that provides direct access to the mapped short reads stored in
the SAM-formatted file, thus minimizing the amount of data that is actually
loaded into memory.
bm = BioMap('s1.sam')

bm =
BioMap with properties:
SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

{1x25 cell}
[458367x1 File
[458367x1 File
[458367x1 File
[458367x1 File
[458367x1 File
[458367x1 File
[458367x1 File
[458367x1 File
[458367x1 File
458367
''

indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed

property]
property]
property]
property]
property]
property]
property]
property]
property]

Use the getSummary method to obtain a list of the existing references and the
actual number of short read mapped to each one. Observe that the order of
the references is equivalent to the previously created cell string chrs.
getSummary(bm)
BioMap summary:
Name:
Container_Type:
Total_Number_of_Sequences:
Number_of_References_in_Dictionary:

gi|224589800|ref|NC_000001.10|

''
'Data is file indexed.'
458367
25

Number_of_Sequences
39037

Genomic_Range
564571 2492

2-47

High-Throughput Sequence Analysis

gi|224589811|ref|NC_000002.11|
gi|224589815|ref|NC_000003.11|
gi|224589816|ref|NC_000004.11|
gi|224589817|ref|NC_000005.9|
gi|224589818|ref|NC_000006.11|
gi|224589819|ref|NC_000007.13|
gi|224589820|ref|NC_000008.10|
gi|224589821|ref|NC_000009.11|
gi|224589801|ref|NC_000010.10|
gi|224589802|ref|NC_000011.9|
gi|224589803|ref|NC_000012.11|
gi|224589804|ref|NC_000013.10|
gi|224589805|ref|NC_000014.8|
gi|224589806|ref|NC_000015.9|
gi|224589807|ref|NC_000016.9|
gi|224589808|ref|NC_000017.10|
gi|224589809|ref|NC_000018.9|
gi|224589810|ref|NC_000019.9|
gi|224589812|ref|NC_000020.10|
gi|224589813|ref|NC_000021.8|
gi|224589814|ref|NC_000022.10|
gi|224589822|ref|NC_000023.10|
gi|224589823|ref|NC_000024.9|
gi|17981852|ref|NC_001807.4|

23102
23788
16273
20875
16743
17022
12199
13988
15707
37506
21714
6078
14644
13199
15423
22089
5986
17690
10026
6119
7366
12939
2819
66035

39107
578280
56044
50342
277774
146474
162668
21790
179281
203411
79745
19335895
19123810
20145084
92212
56680
111538
63006
119233
9421584
16150315
2774622
2711686
12

You can access the alignments, and perform operations like getting counts
and coverage from bm. For more examples of getting read coverage at the
chromosome level, see Exploring Protein-DNA Binding Sites from Paired-End
ChIP-Seq Data.
Determining Digital Gene Expression

Next, you will determine the mapped reads associated with each Ensembl
gene. Because the strings used in the SAM files to denote the reference names
are different to those provided in the annotations, we find a vector with the
reference index for each gene:
geneReference =

2-48

seqmatch(genes.Reference,chrs,'exact',true);

2431
1977
1909
1806
1708
1588
1462
1410
1355
1343
1337
1150
1072
1025
901
810
779
590
629
480
512
1545
590

Identifying Differentially Expressed Genes from RNA-Seq Data

For each gene, count the mapped reads that overlap any part of the gene.
The read counts for each gene are the digital gene expression of that gene.
Use the getCounts method of a BioMap to compute the read count within a
specified range.

counts = getCounts(bm,genes.Start,genes.Stop,1:genes.NumEntries,geneReferen

Gene expression levels can be best respresented by a table, with each row
representing a gene. Create a table with two columns, set the first column to
the gene symbols and second column to the counts of the first sample.

filenames = {'s1.sam','s2.sam','s3.sam','s4.sam','s5.sam','s6.sam','s8.sam'
samples = {'Mock_1','Mock_2','Mock_3','Mock_4','DHT_1','DHT_2','DHT_3'};
lncap = table(genes.Feature,counts,'VariableNames',{'Gene',samples{1}});

Display the counts for the first ten genes.

lncap(1:10,:)

ans =
Gene
____________

Mock_1
______

'DSTYK'
'KCNJ2'
'DPF3'
'KRT78'
'GPR19'
'SOX9'
'C17orf63'
'AL929472.1'
'INPP5B'
'NME4'

21
1
2
0
1
8
13
0
19
10

Determine the number of genes that have counts greater than or equal to
50 in chromosome 1.

2-49

High-Throughput Sequence Analysis

lichr1 = geneReference == 1; % logical index to genes in chromosome 1

sum(lncap.Mock_1>=50 & lichr1)

ans =
188

Repeat this step for the other six samples (SAM files) in the data set to get
their gene counts and copy the information to the previously created table.

for i = 2:7
bm = BioMap(filenames{i});
counts = getCounts(bm,genes.Start,genes.Stop,1:genes.NumEntries,geneRef
lncap.(samples{i}) = counts;
end

Inspect the first 10 rows in the table with the counts for all seven samples.
lncap(1:10, :)

ans =

2-50

Gene
____________

Mock_1
______

Mock_2
______

Mock_3
______

Mock_4
______

DHT_1
_____

DHT_2
_____

'DSTYK'
'KCNJ2'
'DPF3'
'KRT78'
'GPR19'
'SOX9'
'C17orf63'
'AL929472.1'
'INPP5B'
'NME4'

21
1
2
0
1
8
13
0
19
10

15
0
2
0
2
13
12
0
23
11

15
2
2
0
1
19
16
0
27
14

24
0
2
0
1
15
24
1
24
22

24
0
2
0
0
27
19
0
35
11

24
2
1
0
0
22
12
0
32
20

Identifying Differentially Expressed Genes from RNA-Seq Data

DHT_3
_____
15
2
1
0
0
11
9
0
9
8

The table lncap contains counts for samples from two biological conditions:
mock-treated (Aidx) and DHT-treated (Bidx).
Aidx = logical([1 1 1 1 0 0 0]);
Bidx = logical([0 0 0 0 1 1 1]);

You can plot the counts for a chromosome along the chromosome genome
coordinate. For example, plot the counts for chromosome 1 for mock-treated
sample Mock_1 and DHT-treated sample DHT_1. Add the ideogram for
chromosome 1 to the plot using the chromosomeplot function.
ichr1 = find(lichr1); % linear index to genes in chromosome 1
[~,h] = sort(genes.Start(ichr1));
ichr1 = ichr1(h);
% linear index to genes in chromosome 1 sorted by
% genomic position
figure
plot(genes.Start(ichr1), lncap{ichr1,'Mock_1'}, '.-r',...
genes.Start(ichr1), lncap{ichr1,'DHT_1'}, '.-b');
ylabel('Gene Counts')
title('Gene Counts on Chromosome 1')
fixGenomicPositionLabels(gca) % formats tick labels and adds datacursors
chromosomeplot('hs_cytoBand.txt', 1, 'AddToPlot', gca)

2-51

High-Throughput Sequence Analysis

Inference of Differential Signal in RNA Expression

For RNA-seq experiments, the read counts have been found to be linearly
related to the abundance of the target transcripts [4]. The interest lies
in comparing the read counts between different biological conditions.
Current observations suggest that typical RNA-seq experiments have low
background noise, and the gene counts are discrete and could follow the
Poisson distribution. While it has been noted that the assumption of the
Poisson distribution often predicts smaller variation in count data by ignoring
the extra variation due to the actual differences between replicate samples
[5]. Anders et.al.,(2010) proposed an error model for statistical inference
of differential signal in RNA-seq expression data that could address the
overdispersion problem [6]. Their approach uses the negative binomial
distribution to model the null distribution of the read counts. The mean and
variance of the negative binomial distribution are linked by local regression,
and these two parameters can be reliably estimated even when the number of
replicates is small [6].
In this example, you will apply this statistical model to process the count
data and test for differential expression. The details of the algorithm can be
found in reference [6]. The model of Anders et.al., (2010) has three sets of
parameters that need to be estimated from the data:
1. Library size parameters;
2. Gene abundance parameters under each experimental condition;
3. The smooth functions that model the dependence of the raw variance on
the expected mean.
Estimating Library Size Factor

The expectation values of all gene counts from a sample are proportional to
the samples library size. The effective library size can be estimated from the
count data.
Compute the geometric mean of the gene counts (rows in lncap) across all
samples in the experiment as a pseudo-reference sample.
pseudo_ref_sample = geomean(lncap{:,samples},2);

2-52

Identifying Differentially Expressed Genes from RNA-Seq Data

Each library size parameter is computed as the median of the ratio of the
samples counts to those of the pseudo-reference sample.
nzi = pseudo_ref_sample>0; % ignore genes with zero geometric mean
ratios = bsxfun(@rdivide, lncap{nzi,samples}, pseudo_ref_sample(nzi));
sizeFactors = median(ratios, 1);

The counts can be transformed to a common scale using size factor adjustment.
base_lncap = lncap;
base_lncap{:,samples} = bsxfun(@rdivide,lncap{:,samples},sizeFactors);

Use the boxplot function to inspect the count distribution of the mock-treated
and DHT-treated samples and the size factor adjustment.

figure
subplot(2,1,1)
maboxplot(log2(lncap{:,samples}), 'title','Raw Read Counts',...
'orientation', 'horizontal')
subplot(2,1,2)
maboxplot(log2(base_lncap{:,samples}), 'title','Size Factor Adjusted Read C
'orientation', 'horizontal')

Estimate the gene abundance

To estimate the gene abundance for each experimental condition

(mock-treated (A) and DHT-treated (B)) you use the average of the counts
from the samples transformed to the common scale. (Eq. 6 in [6])
mean_A = mean(base_lncap{:,samples(Aidx)}, 2);
mean_B = mean(base_lncap{:,samples(Bidx)}, 2);

Plot the log2 fold changes against the base means using the mairplot
function. A quick exploration reflects ~15 differentially expressed genes (20
fold change or more), though not all of these are significant due to the low
number of counts compared to the sample variance.
mairplot(mean_A(nzi),mean_B(nzi),'Labels',lncap.Gene,'Factor',20)

2-53

High-Throughput Sequence Analysis

Estimating Negative Binomial Distribution Parameters

In the model, the variances of the counts of a gene are considered as the sum
of a shot noise term and a raw variance term. The shot noise term is the
mean counts of the gene, while the raw variance can be predicted from the
mean, i.e., genes with a similar expression level have similar variance across
the replicates (samples of the same biological condition). A smooth function
that models the dependence of the raw variance on the mean is obtained by
fitting the sample mean and variance within replicates for each gene using
local regression function.
Compute sample variances transformed to the common scale for mock-treated
samples. (Eq. 7 in [6])
var_A = var(base_lncap{:,samples(Aidx)}, 0, 2);

Estimate the shot noise term. (Eq. 8 in [6])

z = mean_A * mean(1./sizeFactors(Aidx));

The helper function estimateNBVarFunc returns an anonymous function

that maps the mean estimate to an unbiased raw variance estimate. Bias
adjustment due to shot noise and multiple replicates is considered in the
anonymous function.
raw_var_func_A = estimateNBVarFunc(mean_A,var_A,sizeFactors(Aidx))

raw_var_func_A =
@(meanEstimate)calculateUnbiasedRawVariance(meanEstimate)

Use the anonymous function raw_var_func_A to calculate the sample

variance by adding the shot noise bias term to the raw variance. (Eq.9 in [6])
var_fit_A =

2-54

raw_var_func_A(mean_A) + z;

Identifying Differentially Expressed Genes from RNA-Seq Data

Plot the sample variance to its regressed value to check the fit of the variance
function.
figure
loglog(mean_A, var_A, '*')
hold on
loglog(mean_A, var_fit_A, '.r')
ylabel('Base Variances')
xlabel('Base Means')
title('Dependence of the Variance on the Mean for Mock-Treated Samples')

The fit (red line) follows the single-gene estimates well, even though the
spread of the latter is considerable, as one would expect, given that each
raw variance value is estimated from only four values (four mock-treaded
replicates).
Empirical Cumulative Distribution Functions

As RNA-seq experiments typically have few replicates, the single-gene

estimate of the base variance can deviate wildly from the fitted value. To see
whether this might be too wild, the cumulative probability for the ratio of
single-gene estimate of the base variance to the fitted value is calculated from
the chi-square distribution, as explained in reference [6].
Compute the cumulative probabilities of the variance ratios of mock-treated
samples.
degrees_of_freedom = sum(Aidx) - 1;
var_ratio = var_A ./ var_fit_A;
pchisq = chi2cdf(degrees_of_freedom * var_ratio, degrees_of_freedom);

Compute the empirical cumulative density functions (ECDF) stratified by base

count levels, and show the ECDFs curves. Group the counts into seven levels.
count_levels = [0 3 12 30 65 130 310];
labels = {'0-3','4-12','13-30','31-65','66-130','131-310','> 311'};
grps = sum(bsxfun(@ge,mean_A,count_levels),2); % stratification
figure;

2-55

High-Throughput Sequence Analysis

hold on
cm = jet(7);
for i = 1:7
[Y1,X1] = ecdf(pchisq(grps==i));
plot(X1,Y1,'LineWidth',2,'color',cm(i,:))
end
plot([0,1],[0,1] ,'k', 'linewidth', 2)
set(gca, 'Box', 'on')
legend(labels,'Location','NorthWest')
xlabel('Chi-squared probability of residual')
ylabel('ECDF')
title('Residuals ECDF plot for mock-treated samples')

The ECDF curves of count levels greater than 3 and below 130 follows the
diagonal well (black line). If the ECDF curves are below the black line,
variance is underestimated. If the ECDF curves are above the black line,
variance is overestimated [6]. For very low counts (below 3), the deviations
become stronger, but at these levels, shot noise dominates. For the high
count cases, the variance is overestimated. The reason might be there are
not enough genes with high counts. Get the number of genes in each of the
count levels.

array2table(accumarray(grps,1),'VariableNames',{'Counts'},'RowNames',labels

ans =
Counts
______
0-3
4-12
13-30
31-65
66-130
131-310
> 311

2-56

8984
3405
3481
2418
1173
428
123

Identifying Differentially Expressed Genes from RNA-Seq Data

Increasing the sequence depth, which in turn increases the number of genes
with higher counts, improves the variance estimation.
Testing for Differential Expression

Having estimated and verified the mean-variance dependence, you can test
for differentially expressed genes between the samples from the mock- and
DHT- treated conditions. Define, as test statistic, the total counts in each
condition, k_A and k_B:
k_A = sum(lncap{:, samples(Aidx)}, 2);
k_B = sum(lncap{:, samples(Bidx)}, 2);

Parameters of the new negative binomial distributions for count sums k_A can
be calculated by Eqs. 12-14 in [6]:

pooled_mean = mean(lncap{:, samples},2);

mean_k_A = pooled_mean * sum(sizeFactors(Aidx));
var_k_A = mean_k_A + raw_var_func_A(pooled_mean) * sum(sizeFactors(Aidx).^2

Repeat the same process for k_B:

var_B = var(base_lncap{:,samples(Bidx)}, 0, 2);

raw_var_func_B = estimateNBVarFunc(mean_B,var_B, sizeFactors(Bidx));
mean_k_B = pooled_mean *sum(sizeFactors(Bidx));
var_k_B = mean_k_B + raw_var_func_B(pooled_mean) * sum(sizeFactors(Bidx).^2

Compute the p-values for the statistical significance of the change from
DHT-treated condition to mock-treated condition. The helper function
computePVal implements the numerical computation of the p-values
presented in the reference [6].
res = table(genes.Feature,'VariableNames',{'Gene'});
res.pvals = computePVal(k_B, mean_k_B, var_k_B, k_A, mean_k_A, var_k_A);

You can empirically adjust the p-values from the multiple tests for false
discovery rate (FDR) with the Benjamini-Hochberg procedure [7] using the
mafdr function.
res.p_fdr = mafdr(res.pvals, 'BHFDR', true);

2-57

High-Throughput Sequence Analysis

Determine the fold change estimated from the DHT-treated to the

mock-treated condition.
fold_change = mean_B ./ mean_A;

Determine the base 2 logarithm of the fold change.

res.log2_fold_change = log2(fold_change);

Plot the log2 fold changes against the base means, and color those genes
with p-values.
figure
scatter(log2(pooled_mean), res.log2_fold_change,3,(res.p_fdr).^(.02),'o')
xlabel('log2 Mean')
ylabel('log2 Fold Change')
colormap(flipud(cool(256)))
hc = colorbar;
set(hc,'YTickLabel',num2str((get(hc,'Ytick').^50)','%6.1g'))
title('Fold Change colored by False Discovery Rate (FDR)')

You can identify up- or down- regulated genes for mean base count levels
over 3.
up_idx = find(res.p_fdr < 0.01 & res.log2_fold_change >= 2 & pooled_mean >
numel(up_idx)

ans =
185

down_idx = find(res.p_fdr < 0.01 & res.log2_fold_change <= -2 & pooled_mean

numel(down_idx)

ans =

2-58

Identifying Differentially Expressed Genes from RNA-Seq Data

190

This analysis identified 375 statistically significant (out of 20,012 genes) that
were differentially up- or down- regulated by hormone treatment. You can
sort table res by statistical significant and display the top list.
[~,h] = sort(res.p_fdr);
res(h(1:20),:)

ans =
Gene
_________

pvals
___________

p_fdr
___________

log2_fold_change
________________

'FKBP5'
'NCAPD3'
'CENPN'
'LIFR'
'DHCR24'
'ERRFI1'
'GLYATL2'
'ACSL3'
'ATF3'
'MLPH'
'STEAP4'
'DBI'
'ABCC4'
'KLK2'
'SAT1'
'CAMK2N1'
'JAM3'
'MBOAT2'
'RHOU'
'NNMT'

0
0
6.6707e-300
2.4939e-284
2.0847e-249
9.2602e-246
8.5613e-244
2.6073e-225
1.2368e-193
2.0119e-185
1.7537e-182
3.787e-173
8.5321e-166
2.7911e-163
1.2922e-161
8.8046e-161
4.7333e-151
1.556e-140
1.4157e-138
5.6484e-138

0
0
4.4498e-296
1.2477e-280
8.3437e-246
3.0886e-242
2.4475e-240
6.5221e-222
2.75e-190
4.0263e-182
3.1905e-179
6.3155e-170
1.3134e-162
3.9897e-160
1.724e-158
1.1012e-157
5.5719e-148
1.7299e-137
1.4911e-135
5.6517e-135

5.0449
5.4914
4.8519
4.0734
3.1845
4.0914
3.4522
3.6953
3.368
2.5466
9.9479
2.7759
2.8211
2.9506
2.6687
-4.2901
5.7235
3.285
4.0932
4.3572

References

2-59

High-Throughput Sequence Analysis

[1] Li, H., Lovci, M.T., Kwon, Y-S., Rosenfeld, M.G., Fu, X-D., and Yeo, G.W.
"Determination of Tag Density Required for Digital Transcriptome Analysis:
Application to an Androgen-Sensitive Prostate Cancer Model", PNAS, 105(51),
pp 20179-20184, 2008.
[2] Langmead, B., Trapnell, C., Pop, M., and Salzberg, S.L. "Ultrafast and
Memory-efficient Alignment of Short DNA Sequences to the Human Genome",
Genome Biology, 10:R25, pp 1-10, 2009.
[3] Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N.,
Marth, G., Abecasis, G., Durbin, R. and 1000 Genome Project Data Processing
Subgroup, "The Sequence Alignment/map (SAM) Format and SAMtools",
Bioinformatics, 25, pp 2078-2079, 2009.
[4] Mortazavi, A., Williams, B.A., McCue, K., Schaeffer, L., and Wold, B.
"Mapping and quantifying mammalian transcriptomes by RNA-Seq", Nature
Methods, 5, pp 621-628, 2008.
[5] Robinson, M.D., and Oshlack, A. "A Scaling Normalization method for
differential Expression Analysis of RNA-seq Data", Genome Biology 11:R25,
1-9, 2010.
[6] Anders, S. and Huber W. "Differential Expression Analysis for Sequence
Count Data", Genome Biology, 11:R106, 2010.
[7] Benjamini, Y., and Hochberg, Y. "Controlling the false discovery rate: a
practical and powerful approach to multiple testing", J. Royal Stat. Soc.,
B 57, 289-300, 1995.
Suggest an enhancement for this example.

2-60

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

Exploring Protein-DNA Binding Sites from Paired-End

ChIP-Seq Data
This example shows how to perform a genome-wide analysis of a transcription
factor in the Arabidopsis Thaliana (Thale Cress) model organism.
For enhanced performance, it is recommended that you run this example on a
64-bit platform, because the memory footprint is close to 2 Gb. On a 32-bit
platform, if you receive "Out of memory" errors when running this example,
try increasing the virtual memory (or swap space) of your operating system or
try setting the 3GB switch (32-bit Windows XP only). These techniques are
described in this document.
Introduction

ChIP-Seq is a technology that is used to identify transcription factors that

interact with specific DNA sites. First chromatin immunoprecipitation
enriches DNA-protein complexes using an antibody that binds to a particular
protein of interest. Then, all the resulting fragments are processed using
high-throughput sequencing. Sequencing fragments are mapped back to the
reference genome. By inspecting over-represented regions it is possible to
mark the genomic location of DNA-protein interactions.
In this example, short reads are produced by the paired-end Illumina
platform. Each fragment is reconstructed from two short reads successfully
mapped, with this the exact length of the fragment can be computed. Using
paired-end information from sequence reads maximizes the accuracy of
predicting DNA-protein binding sites.
Data Set

This example explores the paired-end ChIP-Seq data generated by Wang

et.al. [1] using the Illumina platform. The data set has been courteously
submitted to the Gene Expression Omnibus repository with accession number
GSM424618. The unmapped paired-end reads can be obtained from the
NCBI FTP site.
This example assumes that you:

2-61

High-Throughput Sequence Analysis

(1) downloaded the file SRR054715.sra containing the unmapped short read
and converted it to FASTQ formatted files using the NCBI SRA Toolkit.
(2) produced a SAM formatted file by mapping the short reads to the Thale
Cress reference genome, using a mapper such as BWA [2], Bowtie, or SSAHA2
(which is the mapper used by authors of [1]), and,
(3) ordered the SAM formatted file by reference name first, then by genomic
position.
For the published version of this example, 8,655,859 paired-end short reads
are mapped using the BWA mapper [2]. BWA produced a SAM formatted
file (aratha.sam) with 17,311,718 records (8,655,859 x 2). Repetitive hits
were randomly chosen, and only one hit is reported, but with lower mapping
quality. The SAM file was ordered and converted to a BAM formatted file
using SAMtools [3] before being loaded into MATLAB.
The last part of the example also assumes that you downloaded the
reference genome for the Thale Cress model organism (which includes five
chromosomes). Uncomment the following lines of code to download the
reference from the NCBI repository:
%
%
%
%
%

getgenbank('NC_003070','FileFormat','fasta','tofile','ach1.fasta');
getgenbank('NC_003071','FileFormat','fasta','tofile','ach2.fasta');
getgenbank('NC_003074','FileFormat','fasta','tofile','ach3.fasta');
getgenbank('NC_003075','FileFormat','fasta','tofile','ach4.fasta');
getgenbank('NC_003076','FileFormat','fasta','tofile','ach5.fasta');

Creating a MATLAB Interface to a BAM Formatted File

To create local alignments and look at the coverage we need to construct a

BioMap. BioMap has an interface that provides direct access to the mapped
short reads stored in the BAM formatted file, thus minimizing the amount of
data that is actually loaded to the workspace. Create a BioMap to access all
the short reads mapped in the BAM formatted file.
bm = BioMap('aratha.bam')

bm =

2-62

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

BioMap
Properties:
SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

{5x1 cell}
[14637324x1
[14637324x1
[14637324x1
[14637324x1
[14637324x1
[14637324x1
[14637324x1
[14637324x1
[14637324x1
14637324
''

File
File
File
File
File
File
File
File
File

indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed

property]
property]
property]
property]
property]
property]
property]
property]
property]

Use the getSummary method to obtain a list of the existing references and the
actual number of short read mapped to each one.
getSummary(bm)
BioMap summary:
Name:
Container_Type:
Total_Number_of_Sequences:
Number_of_References_in_Dictionary:

Chr1
Chr2
Chr3
Chr4
Chr5

Number_of_Sequences
3151847
3080417
3062917
2218868
3123275

''
'Data is file indexed.'
14637324
5

Genomic_Range
1 30427671
1000 19698292
94 23459782
1029 18585050
11 26975502

2-63

High-Throughput Sequence Analysis

The remainder of this example focuses on the analysis of one of the five
chromosomes, Chr1. Create a new BioMap to access the short reads mapped to
the first chromosome by subsetting the first one.
bm1 = getSubset(bm,'SelectReference','Chr1')

bm1 =
BioMap
Properties:
SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

{'Chr1'}
[3151847x1
[3151847x1
[3151847x1
[3151847x1
[3151847x1
[3151847x1
[3151847x1
[3151847x1
[3151847x1
3151847
''

File
File
File
File
File
File
File
File
File

indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed

property]
property]
property]
property]
property]
property]
property]
property]
property]

By accessing the Start and Stop positions of the mapped short read you can
obtain the genomic range.
x1 = min(getStart(bm1))
x2 = max(getStop(bm1))

x1 =
1

x2 =

2-64

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

30427671

Exploring the Coverage at Different Resolutions

To explore the coverage for the whole range of the chromosome, a binning
algorithm is required. The getBaseCoverage method produces a coverage
signal based on effective alignments. It also allows you to specify a bin width
to control the size (or resolution) of the output signal. However internal
computations are still performed at the base pair (bp) resolution. This means
that despite setting a large bin size, narrow peaks in the coverage signal can
still be observed. Once the coverage signal is plotted you can program the
figures data cursor to display the genomic position when using the tooltip.
You can zoom and pan the figure to determine the position and height of
the ChIP-Seq peaks.
[cov,bin] = getBaseCoverage(bm1,x1,x2,'binWidth',1000,'binType','max');
figure
plot(bin,cov)
axis([x1,x2,0,100])
% sets the axis limits
fixGenomicPositionLabels
% formats tick labels and adds datacursors
xlabel('Base position')
ylabel('Depth')
title('Coverage in Chromosome 1')

It is also possible to explore the coverage signal at the bp resolution (also

referred to as the pile-up profile). Explore one of the large peaks observed in
the data at position 4598837.
p1 = 4598837-1000;
p2 = 4598837+1000;
figure
plot(p1:p2,getBaseCoverage(bm1,p1,p2))
xlim([p1,p2])
% sets the x-axis limits
fixGenomicPositionLabels
% formats tick labels and adds datacursors
xlabel('Base position')

2-65

High-Throughput Sequence Analysis

ylabel('Depth')
title('Coverage in Chromosome 1')

Identifying and Filtering Regions with Artifacts

Observe the large peak with coverage depth of 800+ between positions
4599029 and 4599145. Investigate how these reads are aligning to the
reference chromosome. You can retrieve a subset of these reads enough to
satisfy a coverage depth of 25, since this is sufficient to understand what is
happening in this region. Use getIndex to obtain indices to this subset. Then
use getCompactAlignment to display the corresponding multiple alignment of
the short-reads.
i = getIndex(bm1,4599029,4599145,'depth',25);
bmx = getSubset(bm1,i,'inmemory',false)
getCompactAlignment(bmx,4599029,4599145)

bmx =
BioMap
Properties:
SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

2-66

{'Chr1'}
[62x1 File
[62x1 File
[62x1 File
[62x1 File
[62x1 File
[62x1 File
[62x1 File
[62x1 File
[62x1 File
62
''

indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed

property]
property]
property]
property]
property]
property]
property]
property]
property]

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

ans =

AGTT AATCAAATAGAAAGCCCCGAGGGCGCCATATCCTAGGCGC AAACTATGTGATTGAATAAATCCTCCTC

AGTGC TCAAATAGAAAGCCCCGAGGGCGCCATATTCTAGGAGCCC
GAATAAATCCTCCTC
AGTTCAA
CCCGAGGGCGCCATATTCTAGGAGCCCAAACTATGTGATT
AGTTCAATCAAATAGAAAGC
TTCTAGGAGCCCAAACTATGTGATTGAATAAATCCTCCTC
AGTT
AAGGAGCCCAAAATATGTGATTGAATAAATCCACCTC
AGTACAATCAAATAGAAAGCCCCGAGGGCGCCATA
TAGGAGCCCAAACTATGTGATTGAATAAATCCTCCTC
CGTACAATCAAATAGAAAGCCCCGAGGGCGCCATATTC GGAGCCCAAACTATGTGATTGAATAAATCCTCCTC
CGTACAATCAAATAGAAAGCCCCGAGGGCGCCATATTC GGAGCCCAAACTATGTGATTGAATAAATCCTCCTC
CGTACAATCAAATAGAAAGCCCCGAGGGCGCCATATTC GGAGCCCAAGCTATGTGATTGAATAAATCCTCCTC
CGTACAATCAAATAGAAAGCCCCGAGGGCGCCATATTC GGAGCCCAAACTATGTGATTGAATAAATCCTCCTC
AGTTCAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTA GAGCCCAAACTATGTGATTGAATAAATCCTCCTC
GATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTA GAGCCCAAACTATGTGATTGAATAAATCTTCCTC
GATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTA GAGCCCAAACTATGTGATTGAATAAATCCTCCTC
GATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTA GAGCCCAAACTATGTGATTGAATAAATCCTCCTC
GATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTA GAGCCCAAATTATGTGATTGAATAAATCCTCCTC
ATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAG
CCCAAACTATGTGATTGAATAAATCCTCCTC
ATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAG
CACAAACTATGTGATTGAATAAATCCTCCTC
ATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAG
CCAAACTATGTGATTGAATAAATCCTCCTC
ATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAG
ATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTCG
ATACAATCAAATAGAAAGCCCCGGGGGCGCCATATTCTAG
ATTGAGTCAAATAGAAAGCCCCGAGGGCGCCATATTCTAG
ATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAG
ATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAG
ATACAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAG
CAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAGGAG
CAATCAAATAGAAAGCCCCGAGGGCGCCATATTCTAGGAG
TAGGAGCCCAAACTATGTGATTGAATAAATCCTCCTC
TAGGAGCCCAAACTATGCCATTGAATAAATCCTCCGC
GGAGCCCAAGCTATGTGATTGAATAAATCCTCCTC
GAGCCCAAACTATGTGATTGAATAAATCCTCCTC
GAGCCCAAACTATGTGATTGAATAAATCCTCCTC
GAGCCCAAACTATGTGATTGAATAAATCCTCCTC
GAGCCCAAACTATGTGATTGAATAAATCCTCCTC
GAGCCCAAACTATGTGATTGAATAAATCCTCCTC

2-67

High-Throughput Sequence Analysis

In addition to visually confirming the alignment, you can also explore the
mapping quality for all the short reads in this region, as this may hint to a
potential problem. In this case, less than one percent of the short reads have
a Phred quality of 60, indicating that the mapper most likely found multiple
hits within the reference genome, hence assigning a lower mapping quality.
figure
i = getIndex(bm1,4599029,4599145);
hist(double(getMappingQuality(bm1,i)))
title('Mapping Quality of the reads between 4599029 and 4599145')
xlabel('Phred Quality Score')
ylabel('Number of Reads')

Most of the large peaks in this data set occur due to satellite repeat regions or
due to its closeness to the centromere [4], and show characteristics similar to
the example just explored. You may explore other regions with large peaks
using the same procedure.
To prevent these problematic regions, two techniques are used. First, given
that the provided data set uses paired-end sequencing, by removing the reads
that are not aligned in a proper pair reduces the number of potential aligner
errors or ambiguities. You can achieve this by exploring the flag field of the
SAM formatted file, in which the second less significant bit is used to indicate
if the short read is mapped in a proper pair.
i = find(bitget(getFlag(bm1),2));
bm1_filtered = getSubset(bm1,i)

bm1_filtered =
BioMap
Properties:
SequenceDictionary:
Reference:
Signature:
Start:

2-68

{'Chr1'}
[3040724x1 File indexed property]
[3040724x1 File indexed property]
[3040724x1 File indexed property]

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

[3040724x1
[3040724x1
[3040724x1
[3040724x1
[3040724x1
[3040724x1
3040724
''

File
File
File
File
File
File

indexed
indexed
indexed
indexed
indexed
indexed

property]
property]
property]
property]
property]
property]

Second, consider only uniquely mapped reads. You can detect reads that are
equally mapped to different regions of the reference sequence by looking at
the mapping quality, because BWA assigns a lower mapping quality (less
than 60) to this type of short read.
i = find(getMappingQuality(bm1_filtered)==60);
bm1_filtered = getSubset(bm1_filtered,i)

bm1_filtered =
BioMap
Properties:
SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

{'Chr1'}
[2313252x1
[2313252x1
[2313252x1
[2313252x1
[2313252x1
[2313252x1
[2313252x1
[2313252x1
[2313252x1
2313252
''

File
File
File
File
File
File
File
File
File

indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed
indexed

property]
property]
property]
property]
property]
property]
property]
property]
property]

2-69

High-Throughput Sequence Analysis

Visualize again the filtered data set using both, a coarse resolution with 1000
bp bins for the whole chromosome, and a fine resolution for a small region of
20,000 bp. Most of the large peaks due to artifacts have been removed.

[cov,bin] = getBaseCoverage(bm1_filtered,x1,x2,'binWidth',1000,'binType','m
figure
plot(bin,cov)
axis([x1,x2,0,100])
% sets the axis limits
fixGenomicPositionLabels
% formats tick labels and adds datacursors
xlabel('Base Position')
ylabel('Depth')
title('Coverage in Chromosome 1 after Filtering')
p1 = 24275801-10000;
p2 = 24275801+10000;
figure
plot(p1:p2,getBaseCoverage(bm1_filtered,p1,p2))
xlim([p1,p2])
% sets the x-axis limits
fixGenomicPositionLabels
% formats tick labels and adds datacursors
xlabel('Base Position')
ylabel('Depth')
title('Coverage in Chromosome 1 after Filtering')

Recovering Sequencing Fragments from the Paired-End Reads

In Wangs paper [1] it is hypothesized that paired-end sequencing data has

the potential to increase the accuracy of the identification of chromosome
binding sites of DNA associated proteins because the fragment length can be
derived accurately, while when using single-end sequencing it is necessary
to resort to a statistical approximation of the fragment length, and use it
indistinctly for all putative binding sites.
Use the paired-end reads to reconstruct the sequencing fragments. First,
get the indices for the forward and the reverse reads in each pair. This

2-70

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

information is captured in the fifth bit of the flag field, according to the SAM
file format.
fow_idx = find(~bitget(getFlag(bm1_filtered),5));
rev_idx = find(bitget(getFlag(bm1_filtered),5));

SAM-formatted files use the same header strings to identify pair mates. By
pairing the header strings you can determine how the short reads in BioMap
are paired. To pair the header strings, simply order them in ascending order
and use the sorting indices (hf and hr) to link the unsorted header strings.
[~,hf] = sort(getHeader(bm1_filtered,fow_idx));
[~,hr] = sort(getHeader(bm1_filtered,rev_idx));
mate_idx = zeros(numel(fow_idx),1);
mate_idx(hf) = rev_idx(hr);

Use the resulting fow_idx and mate_idx variables to retrieve pair mates. For
example, retrieve the paired-end reads for the first 10 fragments.
for j = 1:10
disp(getInfo(bm1_filtered, fow_idx(j)))
disp(getInfo(bm1_filtered, mate_idx(j)))
end
SRR054715.sra.6849385
SRR054715.sra.6849385
SRR054715.sra.6992346
SRR054715.sra.6992346
SRR054715.sra.8438570
SRR054715.sra.8438570
SRR054715.sra.1676744
SRR054715.sra.1676744
SRR054715.sra.6820328
SRR054715.sra.6820328
SRR054715.sra.1559757
SRR054715.sra.1559757
SRR054715.sra.5658991
SRR054715.sra.5658991
SRR054715.sra.4625439
SRR054715.sra.4625439

163 20 60 40M AACCCTAAACCTCTGAATCCTTAATCCCTAAATCCCTAA

83 229 60 40M CCTATTTCTTGTGGTTTTCTTTCCTTCACTTAGCTATGG
99 20 60 40M AACCCTAAACCTCTGAATCCTTAATCCCTAAATCCCTAAA
147 239 60 40M GTGGTTTTCTTTCCTTCACTTAGCTATGGATGGTTTAT
163 47 60 40M CTAAATCCCTAAATCTTTAAATCCTACATCCATGAATCC
83 274 60 40M TATCTTCATTTGTTATATTGGATACAAGCTTTGCTACGA
163 67 60 40M ATCCTACATCCATGAATCCCTAAATACCTAATCCCCTAA
83 283 60 40M TTGTTATATTGGATACAAGCTTTGCTACGATCTACATTT
163 73 60 40M CATCCATGAATCCCTAAATACCTAATTCCCTAAACCCGA
83 267 60 40M GTTGGTGTATCTTCATTTGTTATATTGGATACGAGCTTT
163 103 60 40M TAAACCCGAAACCGGTTTCTCTGGTTGAAACTCATTGT
83 311 60 40M GATCTACATTTGGGAATGTGAGTCTCTTATTGTAACCTT
163 103 60 40M CAAACCCGAAACCGGTTTCTCTGGTTGAAACTCATTGT
83 311 60 40M GATCTACATTTGGGAATGTGAGTCTCTTATTGTAACCTT
163 143 60 40M ATATAATGATAATTTTAGCGTTTTTATGCAATTGCTTA
83 347 60 40M CTTAGTGTTGGTTTATCTCAAGAATCTTATTAATTGTTT

2-71

High-Throughput Sequence Analysis

SRR054715.sra.1007474
SRR054715.sra.1007474
SRR054715.sra.7345693
SRR054715.sra.7345693

163 210 60 40M ATTTGAGGTCAATACAAATCCTATTTCTTGTGGTTTGC

83 408 60 40M TATTGTCATTCTTACTCCTTTGTGGAAATGTTTGTTCTA
99 213 60 40M TGAGGTCAATACAAATCCTATTTCTTGTGGTTTTCTTTC
147 393 60 40M TTATTTTTGGACATTTATTGTCATTCTTACTCCTTTGG

Use the paired-end indices to construct a new BioMap with the minimal
information needed to represent the sequencing fragments. First, calculate
the insert sizes.
J = getStop(bm1_filtered, fow_idx);
K = getStart(bm1_filtered, mate_idx);
L = K - J - 1;

Obtain the new signature (or CIGAR string) for each fragment by using the
short read original signatures separated by the appropriate number of skip
CIGAR symbols (N).
n = numel(L);
cigars = cell(n,1);
for i = 1:n
cigars{i} = sprintf('%dN' ,L(i));
end
cigars = strcat( getSignature(bm1_filtered, fow_idx),...
cigars,...
getSignature(bm1_filtered, mate_idx));

Reconstruct the sequences for the fragments by concatenating the respective

sequences of the paired-end short reads.
seqs = strcat( getSequence(bm1_filtered, fow_idx),...
getSequence(bm1_filtered, mate_idx));

Calculate and plot the fragment size distribution.

J = getStart(bm1_filtered,fow_idx);
K = getStop(bm1_filtered,mate_idx);
L = K - J + 1;
figure
hist(double(L),100)
title(sprintf('Fragment Size Distribution\n %d Paired-end Fragments Mapped
xlabel('Fragment Size')

2-72

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

ylabel('Count')

Construct a new BioMap to represent the sequencing fragments. With this,

you will be able explore the coverage signals as well as local alignments of
the fragments.
bm1_fragments = BioMap('Sequence',seqs,'Signature',cigars,'Start',J)

bm1_fragments =
BioMap
Properties:
SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

{0x1 cell}
{0x1 cell}
{1156626x1 cell}
[1156626x1 uint32]
[0x1 uint8]
[0x1 uint16]
[0x1 uint32]
{0x1 cell}
{1156626x1 cell}
{0x1 cell}
1156626
''

Exploring the Coverage Using Fragment Alignments

Compare the coverage signal obtained by using the reconstructed fragments

with the coverage signal obtained by using individual paired-end reads.
Notice that enriched binding sites, represented by peaks, can be better
discriminated from the background signal.

cov_reads = getBaseCoverage(bm1_filtered,x1,x2,'binWidth',1000,'binType','m
[cov_fragments,bin] = getBaseCoverage(bm1_fragments,x1,x2,'binWidth',1000,'

2-73

High-Throughput Sequence Analysis

figure
plot(bin,cov_reads,bin,cov_fragments)
xlim([x1,x2])
% sets the x-axis limits
fixGenomicPositionLabels
% formats tick labels and adds datacursors
xlabel('Base position')
ylabel('Depth')
title('Coverage Comparison')
legend('Short Reads','Fragments')

Perform the same comparison at the bp resolution. In this dataset, Wang

et.al. [1] investigated a basic helix-loop-helix (bHLH) transcription factor.
bHLH proteins typically bind to a consensus sequence called an E-box (with a
CANNTG motif). Use fastaread to load the reference chromosome, search for
the E-box motif in the 3 and 5 directions, and then overlay the motif positions
on the coverage signals. This example works over the region 1-200,000,
however the figure limits are narrowed to a 3000 bp region in order to better
depict the details.
p1 = 1;
p2 = 200000;
cov_reads = getBaseCoverage(bm1_filtered,p1,p2);
[cov_fragments,bin] = getBaseCoverage(bm1_fragments,p1,p2);
chr1 = fastaread('ach1.fasta');
mp1 = regexp(chr1.Sequence(p1:p2),'CA..TG')+3+p1;
mp2 = regexp(chr1.Sequence(p1:p2),'GT..AC')+3+p1;
motifs = [mp1 mp2];
figure
plot(bin,cov_reads,bin,cov_fragments)
hold on
plot([1;1;1]*motifs,[0;max(ylim);NaN],'r')
xlim([111000 114000])
% sets the x-axis limits
fixGenomicPositionLabels
% formats tick labels and adds datacursors
xlabel('Base position')
ylabel('Depth')

2-74

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

title('Coverage Comparison')
legend('Short Reads','Fragments','E-box motif')

Observe that it is not possible to associate each peak in the coverage signals
with an E-box motif. This is because the length of the sequencing fragments is
comparable to the average motif distance, blurring peaks that are close. Plot
the distribution of the distances between the E-box motif sites.
motif_sep = diff(sort(motifs));
figure
hist(motif_sep(motif_sep<500),50)
title('Distance (bp) between adjacent E-box motifs')
xlabel('Distance (bp)')
ylabel('Counts')

Finding Significant Peaks in the Coverage Signal

Use the function mspeaks to perform peak detection with Wavelets denoising
on the coverage signal of the fragment alignments. Filter putative ChIP peaks
using a height filter to remove peaks that are not enriched by the binding
process under consideration.

putative_peaks = mspeaks(bin,cov_fragments,'noiseestimator',20,...
'heightfilter',10,'showplot',true);
hold on
plot([1;1;1]*motifs(motifs>p1 & motifs<p2),[0;max(ylim);NaN],'r')
xlim([111000 114000])
% sets the x-axis limits
fixGenomicPositionLabels
% formats tick labels and adds datacursors
legend('Coverage from Fragments','Wavelet Denoised Coverage','Putative ChIP
xlabel('Base position')
ylabel('Depth')
title('ChIP-Seq Peak Detection')

2-75

High-Throughput Sequence Analysis

Use the knnsearch function to find the closest motif to each one of the
putative peaks. As expected, most of the enriched ChIP peaks are close to an
E-box motif [1]. This reinforces the importance of performing peak detection
at the finest resolution possible (bp resolution) when the expected density of
binding sites is high, as it is in the case of the E-box motif. This example also
illustrates that for this type of analysis, paired-end sequencing should be
considered over single-end sequencing [1].
h = knnsearch(motifs',putative_peaks(:,1));
distance = putative_peaks(:,1)-motifs(h(:))';
figure
hist(distance(abs(distance)<200),50)
title('Distance to Closest E-box Motif for Each Detected Peak')
xlabel('Distance (bp)')
ylabel('Counts')

References

[1] Wang C., Xu J., Zhang D., Wilson Z.A., and Zhang D. "An effective
approach for identification of in vivo protein-DNA binding sites from
paired-end ChIP-Seq data", BMC Bioinformatics, 11:81, Feb 9, 2010.
[2] Li H. and Durbin R. "Fast and accurate short read alignment with
Burrows-Wheeler transform", Bioinformatics, 25, pp 1754-60, 2009.
[3] Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N.,
Marth, G., Abecasis, G., Durbin, R. and 1000 Genome Project Data Processing
Subgroup "The Sequence Alignment/map (SAM) Format and SAMtools",
Bioinformatics, 25, pp 2078-2079, 2009.
[4] Jothi R, Cuddapah S, Barski A, Cui K, Zhao K. "Genome-wide identification
of in vivo protein-DNA binding sites from ChIP-Seq data", Nucleic Acids
Research, 36(16), pp 5221-31, Sep 2008.
[5] Hoofman B.G., and Jones S.J.M. "Genome-wide identification of
DNA-protein interactions using chromatin immunoprecipitation coupled with
flow cell sequencing", Journal of Endocrinology 201, pp 1-13, 2009.

2-76

Exploring Protein-DNA Binding Sites from Paired-End ChIP-Seq Data

[6] Ramsey SA, Knijnenburg TA, Kennedy KA, Zak DE, Gilchrist M, Gold
ES, Johnson CD, Lampano AE, Litvak V, Navarro G, Stolyar T, Aderem A,
Shmulevich I. "Genome-wide histone acetylation data improve prediction of
mammalian transcription factor binding sites", Bioinformatics, 26(17), pp
2071-5, Sep 1, 2010.
Provide feedback for this example.

2-77

High-Throughput Sequence Analysis

Exploring Genome-wide Differences in DNA Methylation

Profiles
This example shows how to perform a genome-wide analysis of DNA
methylation in the human by using genome sequencing.
Note: For enhanced performance, MathWorks recommends that you run this
example on a 64-bit platform, because the memory footprint is close to 2 GB.
On a 32-bit platform, if you receive "Out of memory" errors when running
this example, try increasing the virtual memory (or swap space) of your
operating system or try setting the 3GB switch (32-bit Windows XP only).
These techniques are described in this document.
Introduction

DNA methylation is an epigenetic modification that modulates gene

expression and the maintenance of genomic organization in normal and
disease processes. DNA methylation can define different states of the cell, and
it is inheritable during cell replication. Aberrant DNA methylation patterns
have been associated with cancer and tumor suppressor genes.
In this example you will explore the DNA methylation profiles of two
human cancer cells: parental HCT116 colon cancer cells and DICERex5
cells. DICERex5 cells are derived from HCT116 cells after the truncation of
the DICER1 alleles. Serre et al. in [1] proposed to study DNA methylation
profiles by using the MBD2 protein as a methyl CpG binding domain and
subsequently used high-throughput sequencing (HTseq). This technique
is commonly know as MBD-Seq. Short reads for two replicates of the two
samples have been submitted to NCBIs SRA archive by the authors of [1].
There are other technologies available to interrogate DNA methylation status
of CpG sites in combination with HTseq, for example MeDIP-seq or the use of
restriction enzymes. You can also analyze this type of data sets following the
approach presented in this example.
Data Sets

You can obtain the unmapped single-end reads for four sequencing
experiments from the NCBI FTP site. Short reads were produced using
Illuminas Genome Analyzer II. Average insert size is 120 bp, and the length
of short reads is 36 bp.

2-78

Exploring Genome-wide Differences in DNA Methylation Profiles

This example assumes that you:

(1) downloaded the files SRR030222.sra, SRR030223.sra, SRR030224.sra
and SRR030225.sra containing the unmapped short reads for two replicates
of from the DICERex5 sample and two replicates from the HCT116 sample
respectively. Converted them to FASTQ-formatted files using the NCBI SRA
Toolkit.
(2) produced SAM-formatted files by mapping the short reads to the reference
human genome (NCBI Build 37.5) using the Bowtie [2] algorithm. Only
uniquely mapped reads are reported.
(3) compressed the SAM formatted files to BAM and ordered them by
reference name first, then by genomic position by using SAMtools [3].
This example also assumes that you downloaded the reference human genome
(GRCh37.p5). You can use the bowtie-inspect command to reconstruct the
human reference directly from the bowtie indices. Or you may download the
reference from the NCBI repository by uncommenting the following line:
% getgenbank('NC_000009','FileFormat','fasta','tofile','hsch9.fasta');
Creating a MATLAB Interface to the BAM-Formatted Files

To explore the signal coverage of the HCT116 samples you need to construct a
BioMap. BioMap has an interface that provides direct access to the mapped
short reads stored in the BAM-formatted file, thus minimizing the amount
of data that is actually loaded into memory. Use the function baminfo to
obtain a list of the existing references and the actual number of short reads
mapped to each one.
info = baminfo('SRR030224.bam','ScanDictionary',true);
fprintf('%-35s%s\n','Reference','Number of Reads');
for i = 1:numel(info.ScannedDictionary)
fprintf('%-35s%d\n',info.ScannedDictionary{i},...
info.ScannedDictionaryCount(i));
end
Reference
gi|224589800|ref|NC_000001.10|

Number of Reads
205065

2-79

High-Throughput Sequence Analysis

gi|224589811|ref|NC_000002.11|
gi|224589815|ref|NC_000003.11|
gi|224589816|ref|NC_000004.11|
gi|224589817|ref|NC_000005.9|
gi|224589818|ref|NC_000006.11|
gi|224589819|ref|NC_000007.13|
gi|224589820|ref|NC_000008.10|
gi|224589821|ref|NC_000009.11|
gi|224589801|ref|NC_000010.10|
gi|224589802|ref|NC_000011.9|
gi|224589803|ref|NC_000012.11|
gi|224589804|ref|NC_000013.10|
gi|224589805|ref|NC_000014.8|
gi|224589806|ref|NC_000015.9|
gi|224589807|ref|NC_000016.9|
gi|224589808|ref|NC_000017.10|
gi|224589809|ref|NC_000018.9|
gi|224589810|ref|NC_000019.9|
gi|224589812|ref|NC_000020.10|
gi|224589813|ref|NC_000021.8|
gi|224589814|ref|NC_000022.10|
gi|224589822|ref|NC_000023.10|
gi|224589823|ref|NC_000024.9|
gi|17981852|ref|NC_001807.4|
Unmapped

187019
73986
84033
96898
87990
120816
111229
106189
112279
104466
87091
53638
64049
60183
146868
195893
60344
166420
148950
310048
76037
32421
18870
1015
6805842

In this example you will focus on the analysis of chromosome 9. Create a

BioMap for the two HCT116 sample replicates.

bm_hct116_1 = BioMap('SRR030224.bam','SelectRef','gi|224589821|ref|NC_00000
bm_hct116_2 = BioMap('SRR030225.bam','SelectRef','gi|224589821|ref|NC_00000

bm_hct116_1 =
BioMap with properties:
SequenceDictionary: 'gi|224589821|ref|NC_000009.11|'
Reference: [106189x1 File indexed property]
Signature: [106189x1 File indexed property]

2-80

Exploring Genome-wide Differences in DNA Methylation Profiles

Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

[106189x1
[106189x1
[106189x1
[106189x1
[106189x1
[106189x1
[106189x1
106189

File
File
File
File
File
File
File

indexed
indexed
indexed
indexed
indexed
indexed
indexed

property]
property]
property]
property]
property]
property]
property]

bm_hct116_2 =
BioMap with properties:
SequenceDictionary:
Reference:
Signature:
Start:
MappingQuality:
Flag:
MatePosition:
Quality:
Sequence:
Header:
NSeqs:
Name:

'gi|224589821|ref|NC_000009.11|'
[107586x1 File indexed property]
[107586x1 File indexed property]
[107586x1 File indexed property]
[107586x1 File indexed property]
[107586x1 File indexed property]
[107586x1 File indexed property]
[107586x1 File indexed property]
[107586x1 File indexed property]
[107586x1 File indexed property]
107586

Using a binning algorithm provided by the getBaseCoverage method, you can

plot the coverage of both replicates for an initial inspection. For reference,
you can also add the ideogram for the human chromosome 9 to the plot using
the chromosomeplot function.
figure
ha = gca;
hold on
n = 141213431;
% length of chromosome 9
[cov,bin] = getBaseCoverage(bm_hct116_1,1,n,'binWidth',100);
h1 = plot(bin,cov,'b');
% plots the binned coverage of bm_hct116_1
[cov,bin] = getBaseCoverage(bm_hct116_2,1,n,'binWidth',100);
h2 = plot(bin,cov,'g');
% plots the binned coverage of bm_hct116_2

2-81

High-Throughput Sequence Analysis

chromosomeplot('hs_cytoBand.txt', 9, 'AddToPlot', ha) % plots an ideogram a

axis(ha,[1 n 0 100])
% zooms-in the y-axis
fixGenomicPositionLabels(ha) % formats tick labels and adds datacursors
legend([h1 h2],'HCT116-1','HCT116-2','Location','NorthEast')
ylabel('Coverage')
title('Coverage for two replicates of the HCT116 sample')
set(gcf,'Position',max(get(gcf,'Position'),[0 0 900 0])) % resize window

Because short reads represent the methylated regions of the DNA, there
is a correlation between aligned coverage and DNA methylation. Observe
the increased DNA methylation close to the chromosome telomeres; it is
known that there is an association between DNA methylation and the role of
telomeres for maintaining the integrity of the chromosomes. In the coverage
plot you can also see a long gap over the chromosome centromere. This is due
to the repetitive sequences present in the centromere, which prevent us from
aligning short reads to a unique position in this region. For the data sets used
in this example, only about 30% of the short reads were uniquely mapped to
the reference genome.
Correlating CpG Islands and DNA Methylation

DNA methylation normally occurs in CpG dinucleotides. Alteration of the

DNA methylation patterns can lead to transcriptional silencing, especially in
the gene promoter CpG islands. But, it is also known that DNA methylation
can block CTCF binding and can silence miRNA transcription among other
relevant functions. In general, it is expected that mapped reads should
preferably align to CpG rich regions.
Load the human chromosome 9 from the reference file hs37.fasta. For this
example, it is assumed that you recovered the reference from the Bowtie
indices using the bowtie-inspect command; therefore hs37.fasta contains
all the human chromosomes. To load only the chromosome 9 you can use the
option nave-value pair BLOCKREAD with the fastaread function.
chr9 = fastaread('hs37.fasta','blockread',9)

chr9 =

2-82

Exploring Genome-wide Differences in DNA Methylation Profiles

Header: [1x91 char]

Sequence: [1x141213431 char]

Use the cpgisland function to find the CpG clusters. Using the standard
definition for CpG islands [4], 200 or more bp islands with 60% or greater
CpGobserved/CpGexpected ratio, leads to 1682 GpG islands found in
chromosome 9.
cpgi = cpgisland(chr9.Sequence)

cpgi =
Starts: [1x1682 double]
Stops: [1x1682 double]

Use the getCounts method to calculate the ratio of aligned bases that are
inside CpG islands. For the first replicate of the sample HCT116, the ratio
is close to 45%.

aligned_bases_in_CpG_islands = getCounts(bm_hct116_1,cpgi.Starts,cpgi.Stops
aligned_bases_total = getCounts(bm_hct116_1,1,n,'method','sum')
ratio = aligned_bases_in_CpG_islands ./ aligned_bases_total

aligned_bases_in_CpG_islands =
1724363

aligned_bases_total =
3822804

ratio =

2-83

High-Throughput Sequence Analysis

0.4511

You can explore high resolution coverage plots of the two sample replicates
and observe how the signal correlates with the CpG islands. For example,
explore the region between 23,820,000 and 23,830,000 bp. This is the 5 region
of the human gene ELAVL2.
r1 = 23820001; % set the region limits
r2 = 23830000;
fhELAVL2 = figure; % keep the figure handle to use it later
hold on
% plot high-resolution coverage of bm_hct116_1
h1 = plot(r1:r2,getBaseCoverage(bm_hct116_1,r1,r2,'binWidth',1),'b');
% plot high-resolution coverage of bm_hct116_2
h2 = plot(r1:r2,getBaseCoverage(bm_hct116_2,r1,r2,'binWidth',1),'g');

% mark the CpG islands within the [r1 r2] region

for i=1:numel(cpgi.Starts)
if cpgi.Starts(i)>r1 && cpgi.Stops(i)<r2 % is CpG island inside [r1 r2]?
px = [cpgi.Starts([i i]) cpgi.Stops([i i])]; % x-coordinates for patc
py = [0 max(ylim) max(ylim) 0];
% y-coordinates for patc
hp = patch(px,py,'r','FaceAlpha',.1,'EdgeColor','r','Tag','cpgi');
end
end
axis([r1 r2 0 20])
% zooms-in the y-axis
fixGenomicPositionLabels(gca) % formats tick labels and adds datacursors
legend([h1 h2 hp],'HCT116-1','HCT116-2','CpG Islands')
ylabel('Coverage')
xlabel('Chromosome 9 position')
title('Coverage for two replicates of the HCT116 sample')

Statistical Modelling of Count Data

To find regions that contain more mapped reads than would be expected by
chance, you can follow a similar approach to the one described by Serre et al.

2-84

Exploring Genome-wide Differences in DNA Methylation Profiles

[1]. The number of counts for non-overlapping contiguous 100 bp windows is

statistically modeled.
First, use the getCounts method to count the number of mapped reads
that start at each window. In this example you use a binning approach
that considers only the start position of every mapped read, following the
approach of Serre et al. However, you may also use the OVERLAP and METHOD
name-value pairs in getCounts to compute more accurate statistics. For
instance, to obtain the maximum coverage for each window considering base
pair resolution, set OVERLAP to 1 and METHOD to MAX.
n = numel(chr9.Sequence); % length of chromosome
w = 1:100:n; % windows of 100 bp

counts_1 = getCounts(bm_hct116_1,w,w+99,'independent',true,'overlap','start
counts_2 = getCounts(bm_hct116_2,w,w+99,'independent',true,'overlap','start

First, try to model the counts assuming that all the windows with counts are
biologically significant and therefore from the same distribution. Use the
negative bionomial distribution to fit a model the count data.
nbp = nbinfit(counts_1);

Plot the fitted model over a histogram of the empirical data.

figure
hold on
emphist = histc(counts_1,0:100); % calculate the empirical distribution
bar(0:100,emphist./sum(emphist),'c','grouped') % plot histogram
plot(0:100,nbinpdf(0:100,nbp(1),nbp(2)),'b','linewidth',2); % plot fitted m
axis([0 50 0 .001])
legend('Empirical Distribution','Negative Binomial Fit')
ylabel('Frequency')
xlabel('Counts')
title('Frequency of counts for 100 bp windows (HCT116-1)')

2-85

High-Throughput Sequence Analysis

The poor fitting indicates that the observed distribution may be due to the
mixture of two models, one that represents the background and one that
represents the count data in methylated DNA windows.
A more realistic scenario would be to assume that windows with a small
number of mapped reads are mainly the background (or null model). Serre et
al. assumed that 100-bp windows contaning four or more reads are unlikely
to be generated by chance. To estimate a good approximation to the null
model, you can fit the left body of the emprirical distribution to a truncated
negative binomial distribution. To fit a truncated distribution use the mle
function. First you need to define an anonymous function that defines the
right-truncated version of nbinpdf.
rtnbinpdf = @(x,p1,p2,t) nbinpdf(x,p1,p2) ./ nbincdf(t-1,p1,p2);

Define the fitting function using another anonymous function.

rtnbinfit = @(x,t) mle(x,'pdf',@(x,p1,p2) rtnbinpdf(x,p1,p2,t),'start',nbin

Before fitting the real data, let us assess the fiting procedure with some
sampled data from a known distribution.
nbp = [0.5 0.2];
% Known coefficients
x = nbinrnd(nbp(1),nbp(2),10000,1); % Random sample
trun = 6;
% Set a truncation threshold

nbphat1 = nbinfit(x);
% Fit non-truncated model to all data
nbphat2 = nbinfit(x(x<trun)); % Fit non-truncated model to truncated data (
nbphat3 = rtnbinfit(x(x<trun),trun); % Fit truncated model to truncated dat
figure
hold on
emphist = histc(x,0:100);
% Calculate the empirical distribution
bar(0:100,emphist./sum(emphist),'c','grouped') % plot histogram
h1 = plot(0:100,nbinpdf(0:100,nbphat1(1),nbphat1(2)),'b-o','linewidth',2);
h2 = plot(0:100,nbinpdf(0:100,nbphat2(1),nbphat2(2)),'r','linewidth',2);
h3 = plot(0:100,nbinpdf(0:100,nbphat3(1),nbphat3(2)),'g','linewidth',2);
axis([0 25 0 .2])
legend([h1 h2 h3],'Neg-binomial fitted to all data',...
'Neg-binomial fitted to truncated data',...
'Truncated neg-binomial fitted to truncated data')

2-86

Exploring Genome-wide Differences in DNA Methylation Profiles

ylabel('Frequency')
xlabel('Counts')

Identifying Significant Methylated Regions

For the two replicates of the HCT116 sample, fit a right-truncated negative
binomial distribution to the observed null model using the rtnbinfit
anonymous function previously defined.
trun = 4; % Set a truncation threshold (as in [1])
pn1 = rtnbinfit(counts_1(counts_1<trun),trun); % Fit to HCT116-1 counts
pn2 = rtnbinfit(counts_2(counts_2<trun),trun); % Fit to HCT116-2 counts

Calculate the p-value for each window to the null distribution.

pval1 = 1 - nbincdf(counts_1,pn1(1),pn1(2));
pval2 = 1 - nbincdf(counts_2,pn2(1),pn2(2));

Calculate the false discovery rate using the mafdr function. Use the
name-value pair BHFDR to use the linear-step up (LSU) procedure ([6]) to
calculate the FDR adjusted p-values. Setting the FDR < 0.01 permits you to
identify the 100-bp windows that are significantly methylated.
fdr1 = mafdr(pval1,'bhfdr',true);
fdr2 = mafdr(pval2,'bhfdr',true);

w1 = fdr1<.01; % logical vector indicating significant windows in HCT116-1

w2 = fdr2<.01; % logical vector indicating significant windows in HCT116-2
w12 = w1 & w2; % logical vector indicating significant windows in both repl
Number_of_sig_windows_HCT116_1 = sum(w1)
Number_of_sig_windows_HCT116_2 = sum(w2)
Number_of_sig_windows_HCT116 = sum(w12)

Number_of_sig_windows_HCT116_1 =
1662

2-87

High-Throughput Sequence Analysis

Number_of_sig_windows_HCT116_2 =
1674

Number_of_sig_windows_HCT116 =
1346

Overall, you identified 1662 and 1674 non-overlapping 100-bp windows in the
two replicates of the HCT116 samples, which indicates there is significant
evidence of DNA methylation. There are 1346 windows that are significant in
both replicates.
For example, looking again in the promoter region of the ELAVL2 human
gene you can observe that in both sample replicates, multiple 100-bp windows
have been marked significant.

figure(fhELAVL2) % bring back to focus the previously plotted figure

plot(w(w1)+50,counts_1(w1),'bs') % plot significant windows in HCT116-1
plot(w(w2)+50,counts_2(w2),'gs') % plot significant windows in HCT116-2
axis([r1 r2 0 100])
title('Significant 100-bp windows in both replicates of the HCT116 sample')

Finding Genes With Significant Methylated Promoter Regions

DNA methylation is involved in the modulation of gene expression. For

instance, it is well known that hypermethylation is associated with the
inactivation of several tumor suppresor genes. You can study in this data set
the methylation of gene promoter regions.
First, download from Ensembl a tab-separated-value (TSV) table with all
protein encoding genes to a text file, ensemblmart_genes_hum37.txt. For
this example, we are using Ensamble release 64. Using Ensembls BioMart
service, you can select a table with the following attributes: chromosome
name, gene biotype, gene name, gene start/end, and strand direction.

2-88

Exploring Genome-wide Differences in DNA Methylation Profiles

Use the provided helper function ensemblmart2gff to convert the downloaded

TSV file to a GFF formatted file. Then use GFFAnnotation to load the file into
MATLAB and create a subset with the genes present in chromosome 9 only.
This results 800 annotated protein-coding genes in the Ensembl database.
GFFfilename = ensemblmart2gff('ensemblmart_genes_hum37.txt');
a = GFFAnnotation(GFFfilename)
a9 = getSubset(a,'reference','9')
numGenes = a9.NumEntries

a =
GFFAnnotation with properties:
FieldNames: {1x9 cell}
NumEntries: 21184

a9 =
GFFAnnotation with properties:
FieldNames: {1x9 cell}
NumEntries: 800

numGenes =
800

Find the promoter regions for each gene. In this example we consider the
proximal promoter as the -500/100 upstream region.
downstream = 500;
upstream
= 100;
geneDir = strcmp(a9.Strand,'+');

% logical vector indicating strands in th

2-89

High-Throughput Sequence Analysis

% calculate promoter's start position for genes in the forward direction

promoterStart(geneDir) = a9.Start(geneDir) - downstream;
% calculate promoter's end position for genes in the forward direction
promoterStop(geneDir) = a9.Start(geneDir) + upstream;
% calculate promoter's start position for genes in the reverse direction
promoterStart(~geneDir) = a9.Stop(~geneDir) - upstream;
% calculate promoter's end position for genes in the reverse direction
promoterStop(~geneDir) = a9.Stop(~geneDir) + downstream;

Use a dataset as a container for the promoter information, as we can later

add new columns to store gene counts and p-values.
promoters = dataset({a9.Feature,'Gene'});
promoters.Strand = char(a9.Strand);
promoters.Start = promoterStart';
promoters.Stop = promoterStop';

Find genes with significant DNA methylation in the promoter region by

looking at the number of mapped short reads that overlap at least one base
pair in the defined promoter region.

promoters.Counts_1 = getCounts(bm_hct116_1,promoters.Start,promoters.Stop,.
'overlap',1,'independent',true);
promoters.Counts_2 = getCounts(bm_hct116_2,promoters.Start,promoters.Stop,.
'overlap',1,'independent',true);

Fit a null distribution for each sample replicate and compute the p-values:

trun = 5; % Set a truncation threshold

pn1 = rtnbinfit(promoters.Counts_1(promoters.Counts_1<trun),trun); % Fit to
pn2 = rtnbinfit(promoters.Counts_2(promoters.Counts_2<trun),trun); % Fit to
promoters.pval_1 = 1 - nbincdf(promoters.Counts_1,pn1(1),pn1(2)); % p-value
promoters.pval_2 = 1 - nbincdf(promoters.Counts_2,pn2(1),pn2(2)); % p-value
Number_of_sig_promoters =

sum(promoters.pval_1<.01 & promoters.pval_2<.01)

Ratio_of_sig_methylated_promoters = Number_of_sig_promoters./numGenes

Number_of_sig_promoters =

2-90

Exploring Genome-wide Differences in DNA Methylation Profiles

74

Ratio_of_sig_methylated_promoters =
0.0925

Observe that only 74 (out of 800) genes in chromosome 9 have significantly

DNA methylated regions (pval<0.01 in both replicates). Display a report of
the 30 genes with the most significant methylated promoter regions.
[~,order] = sort(promoters.pval_1.*promoters.pval_2);
promoters(order(1:30),[1 2 3 4 5 7 6 8])

ans =
Gene
'DMRT3'
'CNTFR'
'GABBR2'
'CACNA1B'
'BARX1'
'FAM78A'
'FOXB2'
'TLE4'
'ASTN2'
'FOXE1'
'MPDZ'
'PTPRD'
'PALM2-AKAP2'
'FAM69B'
'WNK2'
'IGFBPL1'
'AKAP2'
'C9orf4'
'COL5A1'
'LHX3'

Strand
+
+
+
+
+
+
+
+
+
+
-

Start
976464
34590021
101471379
140771741
96717554
134151834
79634071
82186188
120177248
100615036
13279489
10612623
112542089
139606522
95946698
38424344
112542269
111929471
137533120
139096855

Stop
977064
34590621
101471979
140772341
96718154
134152434
79634671
82186788
120177848
100615636
13280089
10613223
112542689
139607122
95947298
38424944
112542869
111930071
137533720
139097455

Counts_1
223
219
404
454
264
497
163
157
141
149
129
145
134
112
108
110
107
102
84
74

2-91

High-Throughput Sequence Analysis

'OLFM1'
'NPR2'
'DBC1'
'SOHLH1'
'PIP5K1B'
'PRDM12'
'ELAVL2'
'ZFP37'
'RP11-35N6.1'
'DMRT2'

pval_1
6.6613e-16
6.6613e-16
6.6613e-16
6.6613e-16
6.6613e-16
6.6613e-16
1.4e-13
3.5649e-13
4.3566e-12
1.2447e-12
2.8679e-11
2.3279e-12
1.3068e-11
4.1911e-10
7.897e-10
5.7523e-10
9.2538e-10
2.0467e-09
3.6266e-08
1.8171e-07
1.5457e-07
4.8093e-07
1.5082e-06
3.4322e-06
2.0943e-06
5.6364e-06
9.2778e-06

2-92

+
+
+
+
+
+

Counts_2
253
226
400
408
286
499
165
151
163
133
148
127
135
144
125
114
106
96
97
91
69
73
62
67
63
61
62

137966768
35791651
122131645
138591274
71320075
133539481
23826235
115818939
103790491
1049854

pval_2
6.6613e-16
6.6613e-16
6.6613e-16
6.6613e-16
6.6613e-16
6.6613e-16
6.0363e-13
4.7348e-12
8.098e-13
6.7598e-11
7.3683e-12
1.6448e-10
5.0276e-11
1.3295e-11
2.2131e-10
1.1364e-09
3.7513e-09
1.6795e-08
1.4452e-08
3.5644e-08
1.0074e-06
5.4629e-07
2.9575e-06
1.3692e-06
2.5345e-06
3.4518e-06
2.9575e-06

137967368
35792251
122132245
138591874
71320675
133540081
23826835
115819539
103791091
1050454

75
68
61
56
59
53
50
59
60
54

Exploring Genome-wide Differences in DNA Methylation Profiles

2.0943e-06
1.7771e-06
4.7762e-06

47
42
46

3.0746e-05
6.8037e-05
3.6016e-05

Finding Intergenic Regions that are Significantly Methylated

Serre et al. [1] reported that, in these data sets, approximately 90% of the
uniquely mapped reads fall outside the 5 gene promoter regions. Using
a similar approach as before, you can find genes that have intergenic
methylated regions. To compensate for the varying lengths of the genes, you
can use the maximum coverage, computed base-by-base, instead of the raw
number of mapped short reads. Another alternative approach to normalize
the counts by the gene length is to set the METHOD name-value pair to rpkm
in the getCounts function.
intergenic = dataset({a9.Feature,'Gene'});
intergenic.Strand = char(a9.Strand);
intergenic.Start = a9.Start;
intergenic.Stop = a9.Stop;

intergenic.Counts_1 = getCounts(bm_hct116_1,intergenic.Start,intergenic.Sto
'overlap','full','method','max','independent',true);
intergenic.Counts_2 = getCounts(bm_hct116_2,intergenic.Start,intergenic.Sto
'overlap','full','method','max','independent',true);
trun = 10; % Set a truncation threshold
pn1 = rtnbinfit(intergenic.Counts_1(intergenic.Counts_1<trun),trun); % Fit
pn2 = rtnbinfit(intergenic.Counts_2(intergenic.Counts_2<trun),trun); % Fit
intergenic.pval_1 = 1 - nbincdf(intergenic.Counts_1,pn1(1),pn1(2)); % p-val
intergenic.pval_2 = 1 - nbincdf(intergenic.Counts_2,pn2(1),pn2(2)); % p-val
Number_of_sig_genes =

sum(intergenic.pval_1<.01 & intergenic.pval_2<.01)

Ratio_of_sig_methylated_genes = Number_of_sig_genes./numGenes
[~,order] = sort(intergenic.pval_1.*intergenic.pval_2);
intergenic(order(1:30),[1 2 3 4 5 7 6 8])

2-93

High-Throughput Sequence Analysis

Number_of_sig_genes =
62

Ratio_of_sig_methylated_genes =
0.0775

ans =
Gene
'AL772363.1'
'CACNA1B'
'SUSD1'
'C9orf172'
'NR5A1'
'BARX1'
'KCNT1'
'GABBR2'
'FOXB2'
'NDOR1'
'KIAA1045'
'ADAMTSL2'
'PAX5'
'OLFM1'
'PBX3'
'FOXE1'
'MPDZ'
'ASTN2'
'ARRDC1'
'IGFBPL1'
'LHX3'
'PAPPA'
'CNTFR'
'DMRT3'
'TUSC1'
'ELAVL2'
'SMARCA2'

2-94

Strand
+
+
+
+
+
+
+
+
+
+
+
+
+
+

Start
140762377
140772241
114803065
139738867
127243516
96713628
138594031
101050391
79634571
140100119
34957484
136397286
36833272
137967268
128508551
100615536
13105703
119187504
140500106
38408991
139088096
118916083
34551430
976964
25676396
23690102
2015342

Stop
140787022
141019076
114937688
139741797
127269709
96717654
138684992
101471479
79635869
140113813
34984679
136440641
37034476
138013025
128729656
100618986
13279589
120177348
140509812
38424444
139096955
119164601
34590121
991731
25678856
23826335
2193624

Counts_1
106
106
88
99
86
77
58
65
51
54
50
55
48
55
45
49
51
43
49
45
44
44
41
40
46
35
36

Exploring Genome-wide Differences in DNA Methylation Profiles

'GAS1'
'GRIN1'
'TLE4'

pval_1
8.3267e-15
8.3267e-15
2.2901e-12
7.4385e-14
4.2677e-12
7.0112e-11
2.5424e-08
2.9078e-09
2.2131e-07
8.7601e-08
3.0134e-07
6.4307e-08
5.585e-07
6.4307e-08
1.4079e-06
4.1027e-07
2.2131e-07
2.6058e-06
4.1027e-07
1.4079e-06
1.9155e-06
1.9155e-06
4.8199e-06
6.5537e-06
1.0346e-06
3.0371e-05
2.2358e-05
4.1245e-05
2.2358e-05
2.2358e-05

+
+

Counts_2
98
98
112
96
90
62
73
58
58
55
55
45
49
42
51
46
42
43
36
39
36
35
37
37
31
41
40
41
38
37

89559279
140032842
82186688

89562104
140063207
82341658

34
36
36

pval_2
1.8097e-14
1.8097e-14
1.1102e-16
3.5083e-14
2.5391e-13
2.5691e-09
6.9018e-11
9.5469e-09
9.5469e-09
2.5525e-08
2.5525e-08
6.7163e-07
1.8188e-07
1.7861e-06
9.4566e-08
4.8461e-07
1.7861e-06
1.2894e-06
1.2564e-05
4.7417e-06
1.2564e-05
1.7377e-05
9.0816e-06
9.0816e-06
6.3417e-05
2.4736e-06
3.4251e-06
2.4736e-06
6.5629e-06
9.0816e-06

For instance, explore the methylation profile of the BARX1 gene, the sixth
significant gene with intergenic methylation in the previous list. The GTF

2-95

High-Throughput Sequence Analysis

formatted file ensemblmart_barx1.gtf contains structural information for

this gene obtained from Ensembl using the BioMart service.
Use GTFAnnotation to load the structural information into MATLAB. There
are two annotated transcripts for this gene.
barx1 = GTFAnnotation('ensemblmart_barx1.gtf')
transcripts = getTranscriptNames(barx1)

barx1 =
GTFAnnotation with properties:
FieldNames: {1x11 cell}
NumEntries: 18

transcripts =
'ENST00000253968'
'ENST00000401724'

Plot the DNA methylation profile for both HCT116 sample replicates with
base-pair resolution. Overlay the CpG islands and plot the exons for each of
the two transcripts along the bottom of the plot.
range = barx1.getRange;
r1 = range(1)-1000; % set the region limits
r2 = range(2)+1000;
figure
hold on
% plot high-resolution coverage of bm_hct116_1
h1 = plot(r1:r2,getBaseCoverage(bm_hct116_1,r1,r2,'binWidth',1),'b');
% plot high-resolution coverage of bm_hct116_2
h2 = plot(r1:r2,getBaseCoverage(bm_hct116_2,r1,r2,'binWidth',1),'g');
% mark the CpG islands within the [r1 r2] region
for i=1:numel(cpgi.Starts)

2-96

Exploring Genome-wide Differences in DNA Methylation Profiles

if cpgi.Starts(i)>r1 && cpgi.Stops(i)<r2 % is CpG island inside [r1 r2]

px = [cpgi.Starts([i i]) cpgi.Stops([i i])]; % x-coordinates for pa
py = [0 max(ylim) max(ylim) 0];
% y-coordinates for pa
hp = patch(px,py,'r','FaceAlpha',.1,'EdgeColor','r','Tag','cpgi');
end
end

% mark the exons at the bottom of the axes

for i = 1:numel(transcripts)
exons = getSubset(barx1,'Transcript',transcripts{i},'Feature','exon');
for j = 1:exons.NumEntries
px = [exons.Start([j j]);exons.Stop([j j])]'; % x-coordinates for pa
py = [0 1 1 0]-i*2-1;
% y-coordinates for pa
hq = patch(px,py,'b','FaceAlpha',.1,'EdgeColor','b','Tag','exon');
end
end

axis([r1 r2 -numel(transcripts)*2-2 80]) % zooms-in the y-axis

fixGenomicPositionLabels(gca) % formats tick labels and adds datacursors
ylabel('Coverage')
xlabel('Chromosome 9 position')
title('High resolution coverage in the BARX1 gene')
legend([h1 h2 hp hq],'HCT116-1','HCT116-2','CpG Islands','Exons','Location'

Observe the highly methylated region in the 5 promoter region (right-most

CpG island). Recall that for this gene trasciption occurs in the reverse strand.
More interesting, observe the highly methylated regions that overlap the
initiation of each of the two annotated transcripts (two middle CpG islands).
Differential Analysis of Methylation Patterns

In the study by Serre et al. another cell line is also analyzed. New cells
(DICERex5) are derived from the same HCT116 colon cancer cells after
truncating the DICER1 alleles. It has been reported in literature [5] that
there is a localized change of DNA methylation at small number of gene
promoters. In this example, you be find significant 100-bp windows in two
sample replicates of the DICERex5 cells following the same approach as the

2-97

High-Throughput Sequence Analysis

parental HCT116 cells, and then you will search statistically significant
differences between the two cell lines.
The helper function getWindowCounts captures the similar steps to find
windows with significant coverage as before. getWindowCounts returns
vectors with counts, p-values, and false discovery rates for each new replicate.

bm_dicer_1 = BioMap('SRR030222.bam','SelectRef','gi|224589821|ref|NC_000009
bm_dicer_2 = BioMap('SRR030223.bam','SelectRef','gi|224589821|ref|NC_000009
[counts_3,pval3,fdr3] = getWindowCounts(bm_dicer_1,4,w,100);
[counts_4,pval4,fdr4] = getWindowCounts(bm_dicer_2,4,w,100);
w3 = fdr3<.01; % logical vector indicating significant windows in DICERex5_
w4 = fdr4<.01; % logical vector indicating significant windows in DICERex5w34 = w3 & w4; % logical vector indicating significant windows in both repl
Number_of_sig_windows_DICERex5_1 = sum(w3)
Number_of_sig_windows_DICERex5_2 = sum(w4)
Number_of_sig_windows_DICERex5 = sum(w34)

Number_of_sig_windows_DICERex5_1 =
908

Number_of_sig_windows_DICERex5_2 =
1041

Number_of_sig_windows_DICERex5 =
759

To perform a differential analysis you use the 100-bp windows that are
significant in at least one of the samples (either HCT116 or DICERex5).

wd = w34 | w12; % logical vector indicating windows included in the diff. a

counts = [counts_1(wd) counts_2(wd) counts_3(wd) counts_4(wd)];

2-98

Exploring Genome-wide Differences in DNA Methylation Profiles

ws = w(wd); % window start for each row in counts

Use the function manorm to normalize the data. The PERCENTILE name-value
pair lets you filter out windows with very large number of counts while
normalizing, since these windows are mainly due to artifacts, such as
repetitive regions in the reference chromosome.
counts_norm = round(manorm(counts,'percentile',90).*100);

Use the function mattest to perform a two-sample t-test to identify

differentially covered windows from the two different cell lines.

pval = mattest(counts_norm(:,[1 2]),counts_norm(:,[3 4]),'bootstrap',true,.

'showhist',true,'showplot',true);

Create a report with the 25 most significant differentially covered windows.

While creating the report use the helper function findClosestGene to
determine if the window is intergenic, intragenic, or if it is in a proximal
promoter region.
[~,ord] = sort(pval);
fprintf('Window Pos
Type
p-value
for i = 1:25
j = ord(i);
[~,msg] = findClosestGene(a9,[ws(j) ws(j)+99]);
fprintf('%10d %-25s %7.6f%5d%5d %5d%5d\n', ...
ws(j),msg,pval(j),counts_norm(j,:));
end
Window Pos
140311701
139546501
10901
120176801
139914801

Type
Intergenic (EXD3)
Intragenic
Intragenic
Intergenic (ASTN2)
Intergenic (ABCA2)

p-value
0.000022
0.001684
0.002478
0.002531
0.002770

HCT116
13
21
258
266
64

13
21
257
270
63

HCT116

DICERe

DICERex5
104
91
434
155
26

105
93
428
155
25

2-99

High-Throughput Sequence Analysis

126128501
71939501
124461001
140086501
79637201
136470801
140918001
100615901
98221901
138730601
89561701
977401
37002601
139744401
126771301
93922501
94187101
136044401
139611201
139716201

Intergenic (CRB2)
Prox. Promoter (FAM189A2)
Intergenic (DAB2IP)
Intergenic (TPRN)
Intragenic
Intragenic
Intergenic (CACNA1B)
Intergenic (FOXE1)
Intergenic (PTCH1)
Intergenic (CAMSAP1)
Intergenic (GAS1)
Intergenic (DMRT3)
Intergenic (PAX5)
Intergenic (PHPT1)
Intragenic
Intragenic
Intragenic
Intragenic
Intergenic (FAM69B)
Intergenic (C9orf86)

0.002968
0.005178
0.005243
0.006053
0.006998
0.006998
0.007555
0.007758
0.009231
0.009552
0.009618
0.009656
0.009808
0.010087
0.010638
0.010672
0.010696
0.010756
0.010756
0.010946

94
107
77
47
52
52
176
262
26
26
77
236
133
47
43
34
73
39
39
73

93
101
76
42
51
51
169
253
30
21
76
245
127
46
46
34
80
34
34
72

129
0
39
123
32
32
71
123
104
97
6
129
207
32
97
149
6
110
110
136

130
0
37
124
31
31
68
118
99
93
12
124
211
31
93
161
6
105
105
130

Plot the DNA methylation profile for the promoter region of gene FAM189A2,
the most signicant differentially covered promoter region from the previous
list. Overlay the CpG islands and the FAM189A2 gene.
range = getRange(getSubset(a9,'Feature','FAM189A2'));
r1 = range(1)-1000;
r2 = range(2)+1000;
figure
hold on
% plot high-resolution coverage of all replicates
h1 = plot(r1:r2,getBaseCoverage(bm_hct116_1,r1,r2,'binWidth',1),'b');
h2 = plot(r1:r2,getBaseCoverage(bm_hct116_2,r1,r2,'binWidth',1),'g');
h3 = plot(r1:r2,getBaseCoverage(bm_dicer_1,r1,r2,'binWidth',1),'r');
h4 = plot(r1:r2,getBaseCoverage(bm_dicer_2,r1,r2,'binWidth',1),'m');

% mark the CpG islands within the [r1 r2] region

for i=1:numel(cpgi.Starts)
if cpgi.Starts(i)>r1 && cpgi.Stops(i)<r2 % is CpG island inside [r1 r2]
px = [cpgi.Starts([i i]) cpgi.Stops([i i])]; % x-coordinates for pat

2-100

Exploring Genome-wide Differences in DNA Methylation Profiles

py = [0 max(ylim) max(ylim) 0];

% y-coordinates for pat
hp = patch(px,py,'r','FaceAlpha',.1,'EdgeColor','r','Tag','cpgi');
end
end
% mark the gene at the bottom of the axes
px = range([1 1 2 2]);
py = [0 1 1 0]-2;
hq = patch(px,py,'b','FaceAlpha',.1,'EdgeColor','b','Tag','gene');

axis([r1 r1+4000 -4 30]) % zooms-in

fixGenomicPositionLabels(gca) % formats tick labels and adds datacursors
ylabel('Coverage')
xlabel('Chromosome 9 position')
title('DNA Methylation profiles along the promoter region of the FAM189A2 g
legend([h1 h2 h3 h4 hp hq],'HCT116-1','HCT116-2','DICERex5-1','DICERex5-2',

Observe that the CpG islands are clearly unmethylated for both of the
DICERex5 replicates.
References

[1] Serre, D., Lee, B.H., and Ting A.H. "MBD-isolated Genome Sequencing
provides a high-throughput and comprehensive survey of DNA methylation in
the human genome", Nucleic Acids Research, 38(2), pp 391-399, 2010.
[2] Langmead, B., Trapnell, C., Pop, M., and Salzberg, S.L. "Ultrafast and
Memory-efficient Alignment of Short DNA Sequences to the Human Genome",
Genome Biology, 10:R25, pp 1-10, 2009.
[3] Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N.,
Marth, G., Abecasis, G., Durbin, R. and 1000 Genome Project Data Processing
Subgroup "The Sequence Alignment/map (SAM) Format and SAMtools",
Bioinformatics, 25, pp 2078-2079, 2009.
[4] Gardiner-Garden, M and Frommer, M. "CpG islands in vertebrate
genomes", J.Mol.Biol. 196, pp 261-282, 1987.

2-101

High-Throughput Sequence Analysis

[5] Ting, A.H., Suzuki, H., Cope, L., Schuebel, K.E., Lee, B.H., Toyota, M.,
Imai, K., Shinomura, Y., Tokino, T. and Baylin, S.B. "A Requirement for
DICER to Maintain Full Promoter CpG Island % Hypermethylation in Human
Cancer Cells", Cancer Research, 68, 2570, April 15, 2008.
[6] Benjamini, Y., Hochberg, Y., "Controlling the false discovery rate: a
practical and powerful approach to multiple testing", Journal of the Royal
Statistical Society, 57, pp 289-300, 1995.
Provide feedback for this example.

2-102

3
Sequence Analysis
Sequence analysis is the process you use to find information about a nucleotide
or amino acid sequence using computational methods. Common tasks in
sequence analysis are identifying genes, determining the similarity of two
genes, determining the protein coded by a gene, and determining the function
of a gene by finding a similar gene in another organism with a known function.
Exploring a Nucleotide Sequence Using Command Line on page 3-2
Exploring a Nucleotide Sequence Using the Sequence Viewer App on
page 3-22
Explore a Protein Sequence Using the Sequence Viewer App on page 3-33
Sequence Alignment on page 3-38
View and Align Multiple Sequences on page 3-58

Sequence Analysis

Exploring a Nucleotide Sequence Using Command Line

In this section...
Overview of Example on page 3-2
Searching the Web for Sequence Information on page 3-2
Reading Sequence Information from the Web on page 3-5
Determining Nucleotide Composition on page 3-6
Determining Codon Composition on page 3-11
Open Reading Frames on page 3-15
Amino Acid Conversion and Composition on page 3-18

Overview of Example
After sequencing a piece of DNA, one of the first tasks is to investigate the
nucleotide content in the sequence. Starting with a DNA sequence, this
example uses sequence statistics functions to determine mono-, di-, and
trinucleotide content, and to locate open reading frames.

Searching the Web for Sequence Information

The following procedure illustrates how to use the MATLAB Help browser
to search the Web for information. In this example you are interested in
studying the human mitochondrial genome. While many genes that code for
mitochondrial proteins are found in the cell nucleus, the mitochondrial has
genes that code for proteins used to produce energy.
First research information about the human mitochondria and find the
nucleotide sequence for the genome. Next, look at the nucleotide content for
the entire sequence. And finally, determine open reading frames and extract
specific gene sequences.
1 Use the MATLAB Help browser to explore the Web. In the MATLAB

Command Window, type

web('http://www.ncbi.nlm.nih.gov/')

3-2

Exploring a Nucleotide Sequence Using Command Line

A separate browser window opens with the home page for the NCBI Web
site.
2 Search the NCBI Web site for information. For example, to search for the

human mitochondrion genome, from the Search list, select Genome , and in
the Search list, enter mitochondrion homo sapiens.

The NCBI Web search returns a list of links to relevant pages.

3 Select a result page. For example, click the link labeled NC_012920.

The MATLAB Help browser displays the NCBI page for the human
mitochondrial genome.

3-3

3-4

Sequence Analysis

Exploring a Nucleotide Sequence Using Command Line

Reading Sequence Information from the Web

The following procedure illustrates how to find a nucleotide sequence in
a public database and read the sequence information into the MATLAB
environment. Many public databases for nucleotide sequences are
accessible from the Web. The MATLAB Command Window provides an
integrated environment for bringing sequence information into the MATLAB
environment.
The consensus sequence for the human mitochondrial genome has the
GenBank accession number NC_012920. Since the whole GenBank entry is
quite large and you might only be interested in the sequence, you can get
just the sequence information.
1 Get sequence information from a Web database. For example, to retrieve

sequence information for the human mitochondrial genome, in the

MATLAB Command Window, type
mitochondria = getgenbank('NC_012920','SequenceOnly',true)

The getgenbank function retrieves the nucleotide sequence from the

GenBank database and creates a character array.
mitochondria =
GATCACAGGTCTATCACCCTATTAACCACTCACGGGAGCTCTCCATGCAT
TTGGTATTTTCGTCTGGGGGGTGTGCACGCGATAGCATTGCGAGACGCTG
GAGCCGGAGCACCCTATGTCGCAGTATCTGTCTTTGATTCCTGCCTCATT
CTATTATTTATCGCACCTACGTTCAATATTACAGGCGAACATACCTACTA
AAGT . . .
2 If you dont have a Web connection, you can load the data from a MAT file

included with the Bioinformatics Toolbox software, using the command

load mitochondria

The load function loads the sequence mitochondria into the MATLAB
Workspace.
3 Get information about the sequence. Type

whos mitochondria

3-5

Sequence Analysis

Information about the size of the sequence displays in the MATLAB

Command Window.
Name

Size

Bytes

Class

mitochondria

1x16569

33138

char

Attributes

Determining Nucleotide Composition

The following procedure illustrates how to determine the monomers and
dimers, and then visualize data in graphs and bar plots. Sections of a DNA
sequence with a high percent of A+T nucleotides usually indicate intergenic
parts of the sequence, while low A+T and higher G+C nucleotide percentages
indicate possible genes. Many times high CG dinucleotide content is located
before a gene.
After you read a sequence into the MATLAB environment, you can use
the sequence statistics functions to determine if your sequence has the
characteristics of a protein-coding region. This procedure uses the human
mitochondrial genome as an example. See Reading Sequence Information
from the Web on page 3-5.
1 Plot monomer densities and combined monomer densities in a graph. In

the MATLAB Command Window, type

ntdensity(mitochondria)

This graph shows that the genome is A+T rich.

3-6

Exploring a Nucleotide Sequence Using Command Line

2 Count the nucleotides using the basecount function.

basecount(mitochondria)

A list of nucleotide counts is shown for the 5-3 strand.

ans =
A:
C:
G:
T:

5124
5181
2169
4094

3-7

Sequence Analysis

3 Count the nucleotides in the reverse complement of a sequence using the

seqrcomplement function.
basecount(seqrcomplement(mitochondria))

As expected, the nucleotide counts on the reverse complement strand are

complementary to the 5-3 strand.
ans =
A:
C:
G:
T:

4094
2169
5181
5124

4 Use the function basecount with the chart option to visualize the

nucleotide distribution.
figure
basecount(mitochondria,'chart','pie');

A pie chart displays in the MATLAB Figure window.

3-8

Exploring a Nucleotide Sequence Using Command Line

5 Count the dimers in a sequence and display the information in a bar chart.

figure
dimercount(mitochondria,'chart','bar')
ans =
AA:
AC:
AG:
AT:

1604
1495
795
1230

3-9

Sequence Analysis

CA:
CC:
CG:
CT:
GA:
GC:
GG:
GT:
TA:
TC:
TG:
TT:

3-10

1534
1771
435
1440
613
711
425
419
1373
1204
513
1004

Exploring a Nucleotide Sequence Using Command Line

Determining Codon Composition

The following procedure illustrates how to look at codons for the six reading
frames. Trinucleotides (codon) code for an amino acid, and there are 64
possible codons in a nucleotide sequence. Knowing the percent of codons
in your sequence can be helpful when you are comparing with tables for
expected codon usage.
After you read a sequence into the MATLAB environment, you can analyze
the sequence for codon composition. This procedure uses the human

3-11

Sequence Analysis

mitochondria genome as an example. See Reading Sequence Information

from the Web on page 3-5.
1 Count codons in a nucleotide sequence. In the MATLAB Command

Window, type
codoncount(mitochondria)

The codon counts for the first reading frame displays.

AAA
ACA
AGA
ATA
CAA
CCA
CGA
CTA
GAA
GCA
GGA
GTA
TAA
TCA
TGA
TTA

167
137
59
126
146
141
40
175
67
81
36
43
157
125
64
96

AAC
ACC
AGC
ATC
CAC
CCC
CGC
CTC
GAC
GCC
GGC
GTC
TAC
TCC
TGC
TTC

171
191
87
131
145
205
54
142
53
101
47
26
118
116
40
107

AAG
ACG
AGG
ATG
CAG
CCG
CGG
CTG
GAG
GCG
GGG
GTG
TAG
TCG
TGG
TTG

71
42
51
55
68
49
29
74
49
16
23
18
94
37
29
47

AAT
ACT
AGT
ATT
CAT
CCT
CGT
CTT
GAT
GCT
GGT
GTT
TAT
TCT
TGT
TTT

130
153
54
113
148
173
27
101
35
59
28
41
107
103
26
78

2 Count the codons in all six reading frames and plot the results in heat maps.
for frame = 1:3
figure
subplot(2,1,1);
codoncount(mitochondria,'frame',frame,'figure',true,...
'geneticcode','Vertebrate Mitochondrial');
title(sprintf('Codons for frame %d',frame));
subplot(2,1,2);
codoncount(mitochondria,'reverse',true,'frame',frame,...
'figure',true,'geneticcode','Vertebrate Mitochondrial');
title(sprintf('Codons for reverse frame %d',frame));
end

3-12

Exploring a Nucleotide Sequence Using Command Line

Heat maps display all 64 codons in the 6 reading frames.

3-13

3-14

Sequence Analysis

Exploring a Nucleotide Sequence Using Command Line

Open Reading Frames

The following procedure illustrates how to locate the open reading frames
using a specific genetic code. Determining the protein-coding sequence for a
eukaryotic gene can be a difficult task because introns (noncoding sections)
are mixed with exons. However, prokaryotic genes generally do not have
introns and mRNA sequences have the introns removed. Identifying the
start and stop codons for translation determines the protein-coding section,
or open reading frame (ORF), in a sequence. Once you know the ORF for a
gene or mRNA, you can translate a nucleotide sequence to its corresponding
amino acid sequence.

3-15

Sequence Analysis

After you read a sequence into the MATLAB environment, you can analyze
the sequence for open reading frames. This procedure uses the human
mitochondria genome as an example. See Reading Sequence Information
from the Web on page 3-5.
1 Display open reading frames (ORFs) in a nucleotide sequence. In the

MATLAB Command Window, type:

seqshoworfs(mitochondria);

If you compare this output to the genes shown on the NCBI page for
NC_012920, there are fewer genes than expected. This is because vertebrate
mitochondria use a genetic code slightly different from the standard genetic
code. For a list of genetic codes, see the Genetic Code table in the aa2nt
reference page.
2 Display ORFs using the Vertebrate Mitochondrial code.

orfs= seqshoworfs(mitochondria,...
'GeneticCode','Vertebrate Mitochondrial',...
'alternativestart',true);

Notice that there are now two large ORFs on the third reading frame. One
starts at position 4470 and the other starts at 5904. These correspond to
the genes ND2 (NADH dehydrogenase subunit 2 [Homo sapiens] ) and
COX1 (cytochrome c oxidase subunit I) genes.
3 Find the corresponding stop codon. The start and stop positions for ORFs

have the same indices as the start positions in the fields Start and Stop.
ND2Start = 4470;
StartIndex = find(orfs(3).Start == ND2Start)
ND2Stop = orfs(3).Stop(StartIndex)

The stop position displays.

ND2Stop =
5511

3-16

Exploring a Nucleotide Sequence Using Command Line

4 Using the sequence indices for the start and stop of the gene, extract the

subsequence from the sequence.

ND2Seq = mitochondria(ND2Start:ND2Stop)

The subsequence (protein-coding region) is stored in ND2Seq and displayed

on the screen.
attaatcccctggcccaacccgtcatctactctaccatctttgcaggcac
actcatcacagcgctaagctcgcactgattttttacctgagtaggcctag
aaataaacatgctagcttttattccagttctaaccaaaaaaataaaccct
cgttccacagaagctgccatcaagtatttcctcacgcaagcaaccgcatc
cataatccttc . . .
5 Determine the codon distribution.

codoncount (ND2Seq)

The codon count shows a high amount of ACC, ATA, CTA, and ATC.
AAA
ACA
AGA
ATA
CAA
CCA
CGA
CTA
GAA
GCA
GGA
GTA
TAA
TCA
TGA
TTA

10
11
0
23
8
4
0
26
5
8
5
3
0
7
10
8

AAC
ACC
AGC
ATC
CAC
CCC
CGC
CTC
GAC
GCC
GGC
GTC
TAC
TCC
TGC
TTC

14
24
4
24
3
12
3
18
0
7
7
2
8
11
0
7

AAG
ACG
AGG
ATG
CAG
CCG
CGG
CTG
GAG
GCG
GGG
GTG
TAG
TCG
TGG
TTG

2
3
0
1
2
2
0
4
1
1
0
0
0
1
1
1

AAT
ACT
AGT
ATT
CAT
CCT
CGT
CTT
GAT
GCT
GGT
GTT
TAT
TCT
TGT
TTT

6
5
1
8
1
5
1
7
0
4
1
3
2
4
0
8

6 Look up the amino acids for codons ATA, CTA, ACC, and ATC.

aminolookup('code',nt2aa('ATA'))
aminolookup('code',nt2aa('CTA'))

3-17

Sequence Analysis

aminolookup('code',nt2aa('ACC'))
aminolookup('code',nt2aa('ATC'))

The following displays:

Ile
Leu
Thr
Ile

isoleucine
leucine
threonine
isoleucine

Amino Acid Conversion and Composition

The following procedure illustrates how to extract the protein-coding sequence
from a gene sequence and convert it to the amino acid sequence for the
protein. Determining the relative amino acid composition of a protein will
give you a characteristic profile for the protein. Often, this profile is enough
information to identify a protein. Using the amino acid composition, atomic
composition, and molecular weight, you can also search public databases
for similar proteins.
After you locate an open reading frame (ORF) in a gene, you can convert it to
an amino sequence and determine its amino acid composition. This procedure
uses the human mitochondria genome as an example. See Open Reading
Frames on page 3-15.
1 Convert a nucleotide sequence to an amino acid sequence. In this example,

only the protein-coding sequence between the start and stop codons is
converted.
ND2AASeq = nt2aa(ND2Seq,'geneticcode',...
'Vertebrate Mitochondrial')

The sequence is converted using the Vertebrate Mitochondrial genetic

code. Because the property AlternativeStartCodons is set to 'true' by
default, the first codon att is converted to M instead of I.
MNPLAQPVIYSTIFAGTLITALSSHWFFTWVGLEMNMLAFIPVLTKKMNP
RSTEAAIKYFLTQATASMILLMAILFNNMLSGQWTMTNTTNQYSSLMIMM
AMAMKLGMAPFHFWVPEVTQGTPLTSGLLLLTWQKLAPISIMYQISPSLN
VSLLLTLSILSIMAGSWGGLNQTQLRKILAYSSITHMGWMMAVLPYNPNM
TILNLTIYIILTTTAFLLLNLNSSTTTLLLSRTWNKLTWLTPLIPSTLLS

3-18

Exploring a Nucleotide Sequence Using Command Line

LGGLPPLTGFLPKWAIIEEFTKNNSLIIPTIMATITLLNLYFYLRLIYST
SITLLPMSNNVKMKWQFEHTKPTPFLPTLIALTTLLLPISPFMLMIL
2 Compare your conversion with the published conversion in the GenPept

database.
ND2protein = getgenpept('YP_003024027','sequenceonly',true)

The getgenpept function retrieves the published conversion from the NCBI
database and reads it into the MATLAB Workspace.
3 Count the amino acids in the protein sequence.

aacount(ND2AASeq, 'chart','bar')

A bar graph displays. Notice the high content for leucine, threonine and
isoleucine, and also notice the lack of cysteine and aspartic acid.

3-19

Sequence Analysis

4 Determine the atomic composition and molecular weight of the protein.

atomiccomp(ND2AASeq)
molweight (ND2AASeq)

The following displays in the MATLAB Workspace:

ans =
C:
H:
N:
O:

3-20

1818
2882
420
471

Exploring a Nucleotide Sequence Using Command Line

S: 25
ans =
3.8960e+004

If this sequence was unknown, you could use this information to identify
the protein by comparing it with the atomic composition of other proteins
in a database.

3-21

Sequence Analysis

Exploring a Nucleotide Sequence Using the Sequence

Viewer App
In this section...
Overview of the Sequence Viewer on page 3-22
Importing a Sequence into the Sequence Viewer on page 3-22
Viewing Nucleotide Sequence Information on page 3-24
Searching for Words on page 3-26
Exploring Open Reading Frames on page 3-29
Closing the Sequence Viewer on page 3-32

Overview of the Sequence Viewer

The Sequence Viewer integrates many of the sequence functions in the
Bioinformatics Toolbox toolbox. Instead of entering commands in the
MATLAB Command Window, you can select and enter options using the app.

Importing a Sequence into the Sequence Viewer

The first step when analyzing a nucleotide or amino acid sequence is to
import sequence information into the MATLAB environment. The Sequence
Viewer can connect to Web databases such as NCBI and EMBL and read
information into the MATLAB environment.
The following procedure illustrates how to retrieve sequence information from
the NCBI database on the Web. This example uses the GenBank accession
number NM_000520, which is the human gene HEXA that is associated with
Tay-Sachs disease.
1 In the MATLAB Command Window, type

seqviewer

Alternatively, click Sequence Viewer on the Apps tab.

The Sequence Viewer opens without a sequence loaded. Notice that the
panes to the right and bottom are blank.

3-22

Exploring a Nucleotide Sequence Using the Sequence Viewer App

2 To retrieve a sequence from the NCBI database, select File > Download

Sequence from > NCBI.

The Download Sequence from NCBI dialog box opens.

3 In the Enter Sequence box, type an accession number for an NCBI

database entry, for example, NM_000520. Click the Nucleotide option

button, and then click OK.
The MATLAB software accesses the NCBI database on the Web, loads
nucleotide sequence information for the accession number you entered,
and calculates some basic statistics.

3-23

Sequence Analysis

Viewing Nucleotide Sequence Information

After you import a sequence into the Sequence Viewer app, you can read
information stored with the sequence, or you can view graphic representations
for ORFs and CDSs.
1 In the left pane tree, click Comments. The right pane displays general

information about the sequence.

2 Now click Features. The right pane displays NCBI feature information,

including index numbers for a gene and any CDS sequences.

3 Click ORF to show the search results for ORFs in the six reading frames.

3-24

Exploring a Nucleotide Sequence Using the Sequence Viewer App

4 Click Annotated CDS to show the protein coding part of a nucleotide

sequence.

3-25

Sequence Analysis

Searching for Words

The following procedure illustrates how to search for characteristic words and
sequence patterns. You will search for sequence patterns like the TATAA box
and patterns for specific restriction enzymes.
1 Select Sequence > Find Word.
2 In the Find Word dialog box, type a sequence word or pattern, for example,

atg, and then click Find.

3-26

Exploring a Nucleotide Sequence Using the Sequence Viewer App

The Sequence Viewer searches and displays the location of the selected
word.

3-27

3-28

Sequence Analysis

Exploring a Nucleotide Sequence Using the Sequence Viewer App

3 Clear the display by clicking the Clear Word Selection button

on

the toolbar.

Exploring Open Reading Frames

The following procedure illustrates how to identify the protein coding part of a
nucleotide sequence and copy it into a new view. Identifying coding sections
of a nucleotide sequence is a common bioinformatics task. After locating the
coding part of a sequence, you can copy it to a new view, translate it to an
amino acid sequence, and continue with your analysis.
1 In the left pane, click ORF.

The Sequence Viewer displays the ORFs for the six reading frames in
the lower-right pane. Hover the cursor over a frame to display information
about it.

2 Click the longest ORF on reading frame 2.

The ORF is highlighted to indicate the part of the sequence that is selected.

3 Right-click the selected ORF and then select Export to Workspace. In

the Export to MATLAB Workspace dialog box, type a variable name, for
example, NM_000520_ORF_2, then click Export.

3-29

Sequence Analysis

The NM_000520_ORF_2 variable is added to the MATLAB Workspace.

4 Select File > Import from Workspace. Type the name of a variable

with an exported ORF, for example, NM_000520_ORF_2, and then click

Import.
The Sequence Viewer adds a tab at the bottom for the new sequence
while leaving the original sequence open.

3-30

Exploring a Nucleotide Sequence Using the Sequence Viewer App

5 In the left pane, click Full Translation. Select Display > Amino Acid

Residue Display > One Letter Code.

The Sequence Viewer displays the amino acid sequence below the
nucleotide sequence.

3-31

Sequence Analysis

Closing the Sequence Viewer

Close the Sequence Viewer from the MATLAB command line using the
following syntax:
seqviewer('close')

3-32

Explore a Protein Sequence Using the Sequence Viewer App

Explore a Protein Sequence Using the Sequence Viewer

App
In this section...
Overview of the Sequence Viewer on page 3-33
Viewing Amino Acid Sequence Statistics on page 3-33
Closing the Sequence Viewer on page 3-37
References on page 3-37

Overview of the Sequence Viewer

The Sequence Viewer integrates many of the sequence functions in the
Bioinformatics Toolbox toolbox. Instead of entering commands in the
MATLAB Command Window, you can select and enter options using the app.

Viewing Amino Acid Sequence Statistics

The following procedure illustrates how to view an amino acid sequence for
an ORF located in a nucleotide sequence. You can import your own amino
acid sequence, or you can get a protein sequence from the GenBank database.
This example uses the GenBank accession number NP_000511.1, which is the
alpha subunit for a human enzyme associated with Tay-Sachs disease.
1 Select File > Download Sequence from > NCBI.

The Download Sequence from NCBI dialog box opens.

2 In the Enter Sequence box, type an accession number for an NCBI

database entry, for example, NP_000511.1. Click the Protein option

button, and then click OK.

3-33

Sequence Analysis

The Sequence Viewer accesses the NCBI database on the Web and loads
amino acid sequence information for the accession number you entered.

3-34

Explore a Protein Sequence Using the Sequence Viewer App

3 Select Display > Amino Acid Color Scheme, and then select Charge,

Function, Hydrophobicity, Structure, or Taylor. For example, select

Function.
The display colors change to highlight charge information about the amino
acid residues. The following table shows color legends for the amino acid
color schemes.

3-35

Sequence Analysis

Amino Acid Color Scheme

Color Legend

Charge

Acidic Red
Basic Light Blue
Neutral Black

Function

Acidic Red
Basic Light Blue
Hydropobic, nonpolar Black
Polar, uncharged Green

3-36

Explore a Protein Sequence Using the Sequence Viewer App

Amino Acid Color Scheme

Color Legend

Hydrophobicity

Hydrophilic Light Blue

Hydrophobic Black

Structure

Ambivalent Dark Green

External Light Blue
Internal Orange

Taylor

Each amino acid is assigned its own

color, based on the colors proposed by
W.R. Taylor.

Closing the Sequence Viewer

Close the Sequence Viewer from the MATLAB command line using the
following syntax:
seqviewer('close')

References
[1] Taylor, W.R. (1997). Residual colours: a proposal for aminochromography.
Protein Engineering 10, 7, 743746.

3-37

Sequence Analysis

Sequence Alignment
In this section...
Overview of Example on page 3-38
Find a Model Organism to Study on page 3-38
Retrieve Sequence Information from a Public Database on page 3-41
Search a Public Database for Related Genes on page 3-43
Locate Protein Coding Sequences on page 3-45
Compare Amino Acid Sequences on page 3-49

Overview of Example
Determining the similarity between two sequences is a common task in
computational biology. Starting with a nucleotide sequence for a human gene,
this example uses alignment algorithms to locate and verify a corresponding
gene in a model organism.

Find a Model Organism to Study

In this example, you are interested in studying Tay-Sachs disease. Tay-Sachs
is an autosomal recessive disease caused by the absence of the enzyme
beta-hexosaminidase A (Hex A). This enzyme is responsible for the breakdown
of gangliosides (GM2) in brain and nerve cells.
First, research information about Tay-Sachs and the enzyme that is associated
with this disease, then find the nucleotide sequence for the human gene
that codes for the enzyme, and finally find a corresponding gene in another
organism to use as a model for study.
1 Use the MATLAB Help browser to explore the Web. In the MATLAB

Command window, type

web('http://www.ncbi.nlm.nih.gov/books/NBK22250/')

The MATLAB Help browser opens with the Tay-Sachs disease page in the
Genes and Diseases section of the NCBI web site. This section provides a
comprehensive introduction to medical genetics. In particular, this page

3-38

Sequence Alignment

contains an introduction and pictorial representation of the enzyme Hex A

and its role in the metabolism of the lipid GM2 ganglioside.

3-39

3-40

Sequence Analysis

Sequence Alignment

2 After completing your research, you have concluded the following:

The gene HEXA codes for the alpha subunit of the dimer enzyme
hexosaminidase A (Hex A), while the gene HEXB codes for the beta subunit
of the enzyme. A third gene, GM2A, codes for the activator protein GM2.
However, it is a mutation in the gene HEXA that causes Tay-Sachs.

Retrieve Sequence Information from a Public

Database
The following procedure illustrates how to find the nucleotide sequence for a
human gene in a public database and read the sequence information into the
MATLAB environment. Many public databases for nucleotide sequences (for
example, GenBank, EMBL-EBI) are accessible from the Web. The MATLAB
Command Window with the MATLAB Help browser provide an integrated
environment for searching the Web and bringing sequence information into
the MATLAB environment.
After you locate a sequence, you need to move the sequence data into the
MATLAB Workspace.
1 Open the MATLAB Help browser to the NCBI Web site. In the MATLAB

Command Widow, type

web('http://www.ncbi.nlm.nih.gov/')

The MATLAB Help browser window opens with the NCBI home page.
2 Search for the gene you are interested in studying. For example, from the

Search list, select Nucleotide, and in the for box enter Tay-Sachs.

The search returns entries for the genes that code the alpha and beta
subunits of the enzyme hexosaminidase A (Hex A), and the gene that codes
the activator enzyme. The NCBI reference for the human gene HEXA has
accession number NM_000520.

3-41

Sequence Analysis

3 Get sequence data into the MATLAB environment. For example, to get

sequence information for the human gene HEXA, type

3-42

Sequence Alignment

humanHEXA = getgenbank('NM_000520')

Note Blank spaces in GenBank accession numbers use the underline

character. Entering 'NM 00520' returns the wrong entry.
The human gene is loaded into the MATLAB Workspace as a structure.
humanHEXA =

LocusName: 'NM_000520'
LocusSequenceLength: '2255'
LocusNumberofStrands: ''
LocusTopology: 'linear'
LocusMoleculeType: 'mRNA'
LocusGenBankDivision: 'PRI'
LocusModificationDate: '13-AUG-2006'
Definition: 'Homo sapiens hexosaminidase A (alpha polypeptide) (HEXA), mRNA.'
Accession: 'NM_000520'
Version: 'NM_000520.2'
GI: '13128865'
Project: []
Keywords: []
Segment: []
Source: 'Homo sapiens (human)'
SourceOrganism: [4x65 char]
Reference: {1x58 cell}
Comment: [15x67 char]
Features: [74x74 char]
CDS: [1x1 struct]
Sequence: [1x2255 char]
SearchURL: [1x108 char]
RetrieveURL: [1x97 char]

Search a Public Database for Related Genes

The following procedure illustrates how to find the nucleotide sequence for
a mouse gene related to a human gene, and read the sequence information
into the MATLAB environment. The sequence and function of many genes
is conserved during the evolution of species through homologous genes.

3-43

Sequence Analysis

Homologous genes are genes that have a common ancestor and similar
sequences. One goal of searching a public database is to find similar genes.
If you are able to locate a sequence in a database that is similar to your
unknown gene or protein, it is likely that the function and characteristics of
the known and unknown genes are the same.
After finding the nucleotide sequence for a human gene, you can do a BLAST
search or search in the genome of another organism for the corresponding
gene. This procedure uses the mouse genome as an example.
1 Open the MATLAB Help browser to the NCBI Web site. In the MATLAB

Command window, type

web('http://www.ncbi.nlm.nih.gov')
2 Search the nucleotide database for the gene or protein you are interested in

studying. For example, from the Search list, select Nucleotide, and in the
for box enter hexosaminidase A.
The search returns entries for the mouse and human genomes. The NCBI
reference for the mouse gene HEXA has accession number AK080777.

3 Get sequence information for the mouse gene into the MATLAB

environment. Type
mouseHEXA = getgenbank('AK080777')

3-44

Sequence Alignment

The mouse gene sequence is loaded into the MATLAB Workspace as a

structure.
mouseHEXA =
LocusName:
LocusSequenceLength:
LocusNumberofStrands:
LocusTopology:
LocusMoleculeType:
LocusGenBankDivision:
LocusModificationDate:
Definition:
Accession:
Version:
GI:
Project:
Keywords:
Segment:
Source:
SourceOrganism:
Reference:
Comment:
Features:
CDS:
Sequence:
SearchURL:
RetrieveURL:

'AK080777'
'1839'
''
'linear'
'mRNA'
'HTC'
'02-SEP-2005'
[1x150 char]
'AK080777'
'AK080777.1'
'26348756'
[]
'HTC; CAP trapper.'
[]
'Mus musculus (house mouse)'
[4x65 char]
{1x8 cell}
[8x66 char]
[33x74 char]
[1x1 struct]
[1x1839 char]
[1x107 char]
[1x97 char]

Locate Protein Coding Sequences

The following procedure illustrates how to convert a sequence from nucleotides
to amino acids and identify the open reading frames. A nucleotide sequence
includes regulatory sequences before and after the protein coding section. By
analyzing this sequence, you can determine the nucleotides that code for
the amino acids in the final protein.
After you have a list of genes you are interested in studying, you can
determine the protein coding sequences. This procedure uses the human gene
HEXA and mouse gene HEXA as an example.

3-45

Sequence Analysis

1 If you did not retrieve gene data from the Web, you can load example data

from a MAT-file included with the Bioinformatics Toolbox software. In the

MATLAB Command window, type
load hexosaminidase

The structures humanHEXA and mouseHEXA load into the MATLAB

Workspace.
2 Locate open reading frames (ORFs) in the human gene. For example, for

the human gene HEXA, type

humanORFs = seqshoworfs(humanHEXA.Sequence)
seqshoworfs creates the output structure humanORFs. This structure

contains the position of the start and stop codons for all open reading
frames (ORFs) on each reading frame.
humanORFs =
1x3 struct array with fields:
Start
Stop

The Help browser opens displaying the three reading frames with the
ORFs colored blue, red, and green. Notice that the longest ORF is in the
first reading frame.

3-46

Sequence Alignment

3 Locate open reading frames (ORFs) in the mouse gene. Type:

3-47

Sequence Analysis

mouseORFs = seqshoworfs(mouseHEXA.Sequence)
seqshoworfs creates the structure mouseORFS.
mouseORFs =
1x3 struct array with fields:
Start
Stop

The mouse gene shows the longest ORF on the first reading frame.

3-48

Sequence Alignment

Compare Amino Acid Sequences

The following procedure illustrates how to use global and local alignment
functions to compare two amino acid sequences. You could use alignment
functions to look for similarities between two nucleotide sequences, but
alignment functions return more biologically meaningful results when you
are using amino acid sequences.
After you have located the open reading frames on your nucleotide sequences,
you can convert the protein coding sections of the nucleotide sequences to
their corresponding amino acid sequences, and then you can compare them
for similarities.
1 Using the open reading frames identified previously, convert the human

and mouse DNA sequences to the amino acid sequences. Because both the
human and mouse HEXA genes were in the first reading frames (default),
you do not need to indicate which frame. Type
humanProtein = nt2aa(humanHEXA.Sequence);
mouseProtein = nt2aa(mouseHEXA.Sequence);
2 Draw a dot plot comparing the human and mouse amino acid sequences.

Type
seqdotplot(mouseProtein,humanProtein,4,3)
ylabel('Mouse hexosaminidase A (alpha subunit)')
xlabel('Human hexosaminidase A (alpha subunit)')

Dot plots are one of the easiest ways to look for similarity between
sequences. The diagonal line shown below indicates that there may be a
good alignment between the two sequences.

3-49

Sequence Analysis

3 Globally align the two amino acid sequences, using the Needleman-Wunsch

algorithm. Type
[GlobalScore, GlobalAlignment] = nwalign(humanProtein,...
mouseProtein);
showalignment(GlobalAlignment)
showalignment displays the global alignment of the two sequences in

the Help browser. Notice that the calculated identity between the two
sequences is 60%.

3-50

Sequence Alignment

3-51

Sequence Analysis

The alignment is very good between amino acid position 69 and 599, after
which the two sequences appear to be unrelated. Notice that there is a
stop (*) in the sequence at this point. If you shorten the sequences to
include only the amino acids that are in the protein you might get a better
alignment. Include the amino acid positions from the first methionine (M) to
the first stop (*) that occurs after the first methionine.
4 Trim the sequence from the first start amino acid (usually M) to the first

stop (*) and then try alignment again. Find the indices for the stops in
the sequences.
humanStops = find(humanProtein == '*')
humanStops =
41

599

611

713

722

730

mouseStops = find(mouseProtein == '*')

mouseStops =
539

557

574

606

Looking at the amino acid sequence for humanProtein, the first M is at

position 70, and the first stop after that position is actually the second
stop in the sequence (position 599). Looking at the amino acid sequence
for mouseProtein, the first M is at position 11, and the first stop after that
position is the first stop in the sequence (position 557).
5 Truncate the sequences to include only amino acids in the protein and

the stop.
humanProteinORF = humanProtein(70:humanStops(2))
humanProteinORF =
MTSSRLWFSLLLAAAFAGRATALWPWPQNFQTSDQRYVLYPNNFQFQYDV
SSAAQPGCSVLDEAFQRYRDLLFGSGSWPRPYLTGKRHTLEKNVLVVSVV
TPGCNQLPTLESVENYTLTINDDQCLLLSETVWGALRGLETFSQLVWKSA
EGTFFINKTEIEDFPRFPHRGLLLDTSRHYLPLSSILDTLDVMAYNKLNV

3-52

Sequence Alignment

FHWHLVDDPSFPYESFTFPELMRKGSYNPVTHIYTAQDVKEVIEYARLRG
IRVLAEFDTPGHTLSWGPGIPGLLTPCYSGSEPSGTFGPVNPSLNNTYEF
MSTFFLEVSSVFPDFYLHLGGDEVDFTCWKSNPEIQDFMRKKGFGEDFKQ
LESFYIQTLLDIVSSYGKGYVVWQEVFDNKVKIQPDTIIQVWREDIPVNY
MKELELVTKAGFRALLSAPWYLNRISYGPDWKDFYIVEPLAFEGTPEQKA
LVIGGEACMWGEYVDNTNLVPRLWPRAGAVAERLWSNKLTSDLTFAYERL
SHFRCELLRRGVQAQPLNVGFCEQEFEQT*

mouseProteinORF = mouseProtein(11:mouseStops(1))
mouseProteinORF =
MAGCRLWVSLLLAAALACLATALWPWPQYIQTYHRRYTLYPNNFQFRYHV
SSAAQAGCVVLDEAFRRYRNLLFGSGSWPRPSFSNKQQTLGKNILVVSVV
TAECNEFPNLESVENYTLTINDDQCLLASETVWGALRGLETFSQLVWKSA
EGTFFINKTKIKDFPRFPHRGVLLDTSRHYLPLSSILDTLDVMAYNKFNV
FHWHLVDDSSFPYESFTFPELTRKGSFNPVTHIYTAQDVKEVIEYARLRG
IRVLAEFDTPGHTLSWGPGAPGLLTPCYSGSHLSGTFGPVNPSLNSTYDF
MSTLFLEISSVFPDFYLHLGGDEVDFTCWKSNPNIQAFMKKKGFTDFKQL
ESFYIQTLLDIVSDYDKGYVVWQEVFDNKVKVRPDTIIQVWREEMPVEYM
LEMQDITRAGFRALLSAPWYLNRVKYGPDWKDMYKVEPLAFHGTPEQKAL
VIGGEACMWGEYVDSTNLVPRLWPRAGAVAERLWSSNLTTNIDFAFKRLS
HFRCELVRRGIQAQPISVGCCEQEFEQT*
6 Globally align the trimmed amino acid sequences. Type

[GlobalScore_trim, GlobalAlignment_trim] = nwalign(humanProteinORF,...

mouseProteinORF);
showalignment(GlobalAlignment_trim)
showalignment displays the results for the second global alignment. Notice
that the percent identity for the untrimmed sequences is 60% and 84% for
trimmed sequences.

3-53

Sequence Analysis

7 Another way to truncate an amino acid sequence to only those amino acids

in the protein is to first truncate the nucleotide sequence with indices from

3-54

Sequence Alignment

the seqshoworfs function. Remember that the ORF for the human HEXA
gene and the ORF for the mouse HEXA were both on the first reading
frame.
humanORFs = seqshoworfs(humanHEXA.Sequence)
humanORFs =
1x3 struct array with fields:
Start
Stop

mouseORFs = seqshoworfs(mouseHEXA.Sequence)
mouseORFs =
1x3 struct array with fields:
Start
Stop

humanPORF = nt2aa(humanHEXA.Sequence(humanORFs(1).Start(1):...
humanORFs(1).Stop(1)));
mousePORF = nt2aa(mouseHEXA.Sequence(mouseORFs(1).Start(1):...
mouseORFs(1).Stop(1)));
[GlobalScore2, GlobalAlignment2] = nwalign(humanPORF, mousePORF);

Show the alignment in the Help browser.

showalignment(GlobalAlignment2)

The result from first truncating a nucleotide sequence before converting it

to an amino acid sequence is the same as the result from truncating the
amino acid sequence after conversion. See the result in step 6.
An alternative method to working with subsequences is to use a local
alignment function with the nontruncated sequences.

3-55

Sequence Analysis

8 Locally align the two amino acid sequences using a Smith-Waterman

algorithm. Type
[LocalScore, LocalAlignment] = swalign(humanProtein,...
mouseProtein)
LocalScore =
1057
LocalAlignment =
RGDQR-AMTSSRLWFSLLLAAAFAGRATALWPWPQNFQTSDQRYV . . .
|| | ||:: ||| |||||||:| ||||||||| :|| :||: . . .
RGAGRWAMAGCRLWVSLLLAAALACLATALWPWPQYIQTYHRRYT . . .
9 Show the alignment in color.

showalignment(LocalAlignment)

3-56

Sequence Alignment

3-57

Sequence Analysis

View and Align Multiple Sequences

In this section...
Overview of the Sequence Alignment and Phylogenetic Tree Apps on page
3-58
Load Sequence Data and Viewing the Phylogenetic Tree on page 3-58
Select a Subset of Data from the Phylogenetic Tree on page 3-59
Align Multiple Sequences on page 3-61
Adjust Multiple Sequence Alignments Manually on page 3-62
Close the Sequence Alignment App on page 3-65

Overview of the Sequence Alignment and

Phylogenetic Tree Apps
The Sequence Alignment app integrates many sequence and multiple
alignment functions in the toolbox. Instead of entering commands in the
MATLAB Command Window, you can use this app to visually inspect a
multiple alignment and make manual adjustments.
The Phylogenetic Tree app allows you to view, edit, and explore
phylogenetic tree data. It also allows branch pruning, reordering, renaming,
and distance exploring. It can also open or save Newick or ClustalW tree
formatted files.

Load Sequence Data and Viewing the Phylogenetic

Tree
Load unaligned sequence data into the MATLAB environment, and create a
phylogenetic tree.
1 Load sequence data.

load primates.mat
2 Create a phylogenetic tree.

tree = seqlinkage(seqpdist(primates),'single', primates);

3-58

View and Align Multiple Sequences

3 View the phylogenetic tree.

phytreeviewer(tree)

Select a Subset of Data from the Phylogenetic Tree

Select the human and chimp branches.
1 From the toolbar, click the Prune icon.

3-59

Sequence Analysis

2 Click the branches to prune (remove) from the tree. For this example, click

the branch nodes for gorillas, orangutans, and Neanderthals.

3 Export the selected branches to a second tree. Select File > Export to

Workspace, and then select Only Displayed.

4 In the Export to dialog box, enter the name of a variable. For example,

enter tree2, and then click OK.

3-60

View and Align Multiple Sequences

5 Extract sequences from the tree object.

primates2 = primates(seqmatch(get(tree2, 'Leafnames'),{primates.Header}));

Align Multiple Sequences

After selecting a set of related sequences, you can align them and view the
results.
1 Align multiple sequences.

ma = multialign(primates2);
2 View the aligned sequences in the Sequence Alignment app.

seqalignviewer(ma);

The aligned sequences appear as shown below.

3-61

Sequence Analysis

Adjust Multiple Sequence Alignments Manually

Algorithms for aligning multiple sequences do not always produce an optimal
result. By visually inspecting the alignment, you can identify areas that could
use a manual adjustment to improve the alignment.
1 Identify an area where you could improve the alignment.

3-62

View and Align Multiple Sequences

2 Click a letter to select it, and then move the cursor over the red direction

bar. The cursor changes to a hand.

3 Click and drag the sequence to the right to insert a gap. If there is a gap to

the left, you can also move the sequence to the left and eliminate the gap.

3-63

Sequence Analysis

Alternately, to insert a gap, select a character, and then click the Insert
Gap icon on the toolbar or press the spacebar.

Note You cannot delete or add letters to a sequence, but you can add or
delete gaps. If all of the sequences at one alignment position have gaps,
you can delete that column of gaps.

3-64

View and Align Multiple Sequences

4 Continue adding gaps and moving sequences to improve the alignment.

Close the Sequence Alignment App

Close the Sequence Alignment app from the MATLAB command line using
the following syntax:
seqalignviewer('close')

3-65

3-66

Sequence Analysis

4
Microarray Analysis
Managing Gene Expression Data in Objects on page 4-2
Representing Expression Data Values in DataMatrix Objects on page 4-5
Representing Expression Data Values in ExptData Objects on page 4-11
Representing Sample and Feature Metadata in MetaData Objects on
page 4-15
Representing Experiment Information in a MIAME Object on page 4-22
Representing All Data in an ExpressionSet Object on page 4-27
Visualizing Microarray Images on page 4-33
Analyzing Gene Expression Profiles on page 4-57
Detecting DNA Copy Number Alteration in Array-Based CGH Data on
page 4-72
Exploring Gene Expression Data on page 4-85

Microarray Analysis

Managing Gene Expression Data in Objects

Microarray gene expression experiments are complex, containing data and
information from various sources. The data and information from such an
experiment is typically subdivided into four categories:
Measured expression data values
Sample metadata
Microarray feature metadata
Descriptions of experiment methods and conditions
In MATLAB, you can represent all the previous data and information in an
ExpressionSet object, which typically contains the following objects:
One ExptData object containing expression values from a microarray
experiment in one or more DataMatrix objects
One MetaData object containing sample metadata in two dataset arrays
One MetaData object containing feature metadata in two dataset arrays
One MIAME object containing experiment descriptions
The following graphic illustrates a typical ExpressionSet object and its
component objects.

4-2

Managing Gene Expression Data in Objects

4-3

Microarray Analysis

Each element (DataMatrix object) in the ExpressionSet object has an element

name. Also, there is always one DataMatrix object whose element name is
Expressions.
An ExpressionSet object lets you store, manage, and subset the data from a
microarray gene expression experiment. An ExpressionSet object includes
properties and methods that let you access, retrieve, and change data,
metadata, and other information about the microarray experiment. These
properties and methods are useful to view and analyze the data. For a list of
the properties and methods, see ExpressionSet class.
To learn more about constructing and using objects for microarray gene
expression data and information, see:
Representing Expression Data Values in DataMatrix Objects on page 4-5
Representing Expression Data Values in ExptData Objects on page 4-11
Representing Sample and Feature Metadata in MetaData Objects on
page 4-15
Representing Experiment Information in a MIAME Object on page 4-22
Representing All Data in an ExpressionSet Object on page 4-27

4-4

Representing Expression Data Values in DataMatrix Objects

Representing Expression Data Values in DataMatrix

Objects
In this section...
Overview of DataMatrix Objects on page 4-5
Constructing DataMatrix Objects on page 4-6
Getting and Setting Properties of a DataMatrix Object on page 4-7
Accessing Data in DataMatrix Objects on page 4-8

Overview of DataMatrix Objects

The toolbox includes functions, objects, and methods for creating, storing, and
accessing microarray data.
The object constructor function, DataMatrix, lets you create a DataMatrix
object to encapsulate data and metadata (row and column names) from a
microarray experiment. A DataMatrix object stores experimental data in a
matrix, with rows typically corresponding to gene names or probe identifiers,
and columns typically corresponding to sample identifiers. A DataMatrix
object also stores metadata, including the gene names or probe identifiers (as
the row names) and sample identifiers (as the column names).
You can reference microarray expression values in a DataMatrix object the
same way you reference data in a MATLAB array, that is, by using linear or
logical indexing. Alternately, you can reference this experimental data by
gene (probe) identifiers and sample identifiers. Indexing by these identifiers
lets you quickly and conveniently access subsets of the data without having
to maintain additional index arrays.
Many MATLAB operators and arithmetic functions are available to
DataMatrix objects by means of methods. These methods let you modify,
combine, compare, analyze, plot, and access information from DataMatrix
objects. Additionally, you can easily extend the functionality by using
general element-wise functions, dmarrayfun and dmbsxfun, and by manually
accessing the properties of a DataMatrix object.

4-5

Microarray Analysis

Note For tables describing the properties and methods of a DataMatrix

object, see the DataMatrix object reference page.

Constructing DataMatrix Objects

1 Load the MAT-file, provided with the Bioinformatics Toolbox software, that

contains yeast data. This MAT-file includes three variables: yeastvalues,

a 614-by-7 matrix of gene expression data, genes, a cell array of 614
GenBank accession numbers for labeling the rows in yeastvalues,
and times, a 1-by-7 vector of time values for labeling the columns in
yeastvalues.
load filteredyeastdata
2 Create variables to contain a subset of the data, specifically the first five

rows and first four columns of the yeastvalues matrix, the genes cell
array, and the times vector.
yeastvalues = yeastvalues(1:5,1:4);
genes = genes(1:5,:);
times = times(1:4);
3 Import the microarray object package so that the DataMatrix constructor

function will be available.

import bioma.data.*
4 Use the DataMatrix constructor function to create a small DataMatrix

object from the gene expression data in the variables you created in step 2.
dmo = DataMatrix(yeastvalues,genes,times)
dmo =

SS DNA
YAL003W
YAL012W
YAL026C

4-6

0
-0.131
0.305
0.157
0.246

9.5
1.699
0.146
0.175
0.796

11.5
-0.026
-0.129
0.467
0.384

13.5
0.365
-0.444
-0.379
0.981

Representing Expression Data Values in DataMatrix Objects

YAL034C

-0.235

0.487

-0.184

-0.669

Getting and Setting Properties of a DataMatrix

Object
You use the get and set methods to retrieve and set properties of a
DataMatrix object.
1 Use the get method to display the properties of the DataMatrix object, dmo.

get(dmo)
Name:
RowNames:
ColNames:
NRows:
NCols:
NDims:
ElementClass:

''
{5x1 cell}
{'
0' ' 9.5'
5
4
2
'double'

'11.5'

'13.5'}

2 Use the set method to specify a name for the DataMatrix object, dmo.

dmo = set(dmo,'Name','MyDMObject');
3 Use the get method again to display the properties of the DataMatrix

object, dmo.
get(dmo)
Name:
RowNames:
ColNames:
NRows:
NCols:
NDims:
ElementClass:

'MyDMObject'
{5x1 cell}
{'
0' ' 9.5'
5
4
2
'double'

'11.5'

'13.5'}

Note For a description of all properties of a DataMatrix object, see the

DataMatrix object reference page.

4-7

Microarray Analysis

Accessing Data in DataMatrix Objects

DataMatrix objects support the following types of indexing to extract, assign,
and delete data:
Parenthesis ( ) indexing
Dot . indexing

Parentheses () Indexing
Use parenthesis indexing to extract a subset of the data in dmo and assign
it to a new DataMatrix object dmo2:
dmo2 = dmo(1:5,2:3)
dmo2 =
9.5
SS DNA
1.699
YAL003W
0.146
YAL012W
0.175
YAL026C
0.796
YAL034C
0.487

11.5
-0.026
-0.129
0.467
0.384
-0.184

Use parenthesis indexing to extract a subset of the data using row names and
column names, and assign it to a new DataMatrix object dmo3:
dmo3 = dmo({'SS DNA','YAL012W','YAL034C'},'11.5')
dmo3 =

SS DNA
YAL012W
YAL034C

11.5
-0.026
0.467
-0.184

Note If you use a cell array of row names or column names to index into a
DataMatrix object, the names must be unique, even though the row names or
column names within the DataMatrix object are not unique.

4-8

Representing Expression Data Values in DataMatrix Objects

Use parenthesis indexing to assign new data to a subset of the elements in

dmo2:
dmo2({'SS DNA', 'YAL003W'}, 1:2) = [1.700 -0.030; 0.150 -0.130]
dmo2 =

SS DNA
YAL003W
YAL012W
YAL026C
YAL034C

9.5
1.7
0.15
0.175
0.796
0.487

11.5
-0.03
-0.13
0.467
0.384
-0.184

Use parenthesis indexing to delete a subset of the data in dmo2:

dmo2({'SS DNA', 'YAL003W'}, :) = []
dmo2 =

YAL012W
YAL026C
YAL034C

9.5
0.175
0.796
0.487

11.5
0.467
0.384
-0.184

Dot . Indexing
Note In the following examples, notice that when using dot indexing with
DataMatrix objects, you specify all rows or all columns using a colon within
single quotation marks, (':').
Use dot indexing to extract the data from the 11.5 column only of dmo:
timeValues = dmo.(':')('11.5')
timeValues =
-0.0260
-0.1290
0.4670
0.3840
-0.1840

4-9

Microarray Analysis

Use dot indexing to assign new data to a subset of the elements in dmo:
dmo.(1:2)(':') = 7
dmo =
0
SS DNA
YAL003W
YAL012W
YAL026C
YAL034C

7
7
0.157
0.246
-0.235

9.5
7
7
0.175
0.796
0.487

11.5

13.5

7
7
0.467
0.384
-0.184

7
7
-0.379
0.981
-0.669

Use dot indexing to delete an entire variable from dmo:

dmo.YAL034C = []
dmo =
0
SS DNA
YAL003W
YAL012W
YAL026C

7
7
0.157
0.246

9.5
7
7
0.175
0.796

11.5

13.5

7
7
0.467
0.384

7
7
-0.379
0.981

Use dot indexing to delete two columns from dmo:

dmo.(':')(2:3)=[]
dmo =
0
SS DNA
YAL003W
YAL012W
YAL026C

4-10

7
7
0.157
0.246

13.5
7
7
-0.379
0.981

Representing Expression Data Values in ExptData Objects

Representing Expression Data Values in ExptData Objects

In this section...
Overview of ExptData Objects on page 4-11
Constructing ExptData Objects on page 4-12
Using Properties of an ExptData Object on page 4-12
Using Methods of an ExptData Object on page 4-13
References on page 4-14

Overview of ExptData Objects

You can use an ExptData object to store expression values from a microarray
experiment. An ExprData object stores the data values in one or more
DataMatrix objects, each having the same row names (feature names) and
column names (sample names). Each element (DataMatrix object) in the
ExptData object has an element name.
The following illustrates a small DataMatrix object containing expression
values from three samples (columns) and seven features (rows):
A
100001_at
100002_at
100003_at
100004_at
100005_at
100006_at
100007_at

2.26
158.86
68.11
74.32
75.05
80.36
216.64

B
20.14
236.25
105.45
96.68
53.17
42.89
191.32

C
31.66
206.27
82.92
84.87
57.94
77.21
219.48

An ExptData object lets you store, manage, and subset the data values from a
microarray experiment. An ExptData object includes properties and methods
that let you access, retrieve, and change data values from a microarray
experiment. These properties and methods are useful to view and analyze the
data. For a list of the properties and methods, see ExptData class.

4-11

Microarray Analysis

Constructing ExptData Objects

The mouseExprsData.txt file used in this example contains data from
Hovatta et al., 2005.
1 Import the bioma.data package so that the DataMatrix and ExptData

constructor functions are available.

import bioma.data.*
2 Use the DataMatrix constructor function to create a DataMatrix object

from the gene expression data in the mouseExprsData.txt file. This

file contains a table of expression values and metadata (sample and
feature names) from a microarray experiment done using the Affymetrix
MGU74Av2 GeneChip array. There are 26 sample names (A through Z),
and 500 feature names (probe set names).
dmObj = DataMatrix('File', 'mouseExprsData.txt');
3 Use the ExptData constructor function to create an ExptData object from

the DataMatrix object.

EDObj = ExptData(dmObj);
4 Display information about the ExptData object, EDObj.

EDObj
Experiment Data:
500 features, 26 samples
1 elements
Element names: Elmt1

Note For complete information on constructing ExptData objects, see

ExptData class.

Using Properties of an ExptData Object

To access properties of an ExptData object, use the following syntax:
objectname.propertyname

4-12

Representing Expression Data Values in ExptData Objects

For example, to determine the number of elements (DataMatrix objects) in

an ExptData object:
EDObj.NElements
ans =
1

To set properties of an ExptData object, use the following syntax:

objectname.propertyname = propertyvalue

For example, to set the Name property of an ExptData object:

EDObj.Name = 'MyExptDataObject'

Note Property names are case sensitive. For a list and description of all
properties of an ExptData object, see ExptData class.

Using Methods of an ExptData Object

To use methods of an ExptData object, use either of the following syntaxes:
objectname.methodname

or
methodname(objectname)

For example, to retrieve the sample names from an ExptData object:

EDObj.sampleNames
Columns 1 through 9
'A'

'B'

'C'

'D'

'E'

'F'

'G'

'H'

'I'

...

To return the size of an ExptData object:

size(EDObj)

4-13

Microarray Analysis

ans =
500

26

Note For a complete list of methods of an ExptData object, see ExptData

class.

References
[1] Hovatta, I., Tennant, R S., Helton, R., et al. (2005). Glyoxalase 1 and
glutathione reductase 1 regulate anxiety in mice. Nature 438, 662666.

4-14

Representing Sample and Feature Metadata in MetaData Objects

Representing Sample and Feature Metadata in MetaData

Objects
In this section...
Overview of MetaData Objects on page 4-15
Constructing MetaData Objects on page 4-16
Using Properties of a MetaData Object on page 4-19
Using Methods of a MetaData Object on page 4-20

Overview of MetaData Objects

You can store either sample or feature metadata from a microarray gene
expression experiment in a MetaData object. The metadata consists of
variable names, for example, related to either samples or microarray features,
along with descriptions and values for the variables.
A MetaData object stores the metadata in two dataset arrays:
Values dataset array A dataset array containing the measured value
of each variable per sample or feature. In this dataset array, the columns
correspond to variables and rows correspond to either samples or features.
The number and names of the columns in this dataset array must match
the number and names of the rows in the Descriptions dataset array. If
this dataset array contains sample metadata, then the number and names
of the rows (samples) must match the number and names of the columns
in the DataMatrix objects in the same ExpressionSet object. If this
dataset array contains feature metadata, then the number and names of
the rows (features) must match the number and names of the rows in the
DataMatrix objects in the same ExpressionSet object.
Descriptions dataset array A dataset array containing a list of the
variable names and their descriptions. In this dataset array, each row
corresponds to a variable. The row names are the variable names, and
a column, named VariableDescription, contains a description of the
variable. The number and names of the rows in the Descriptions dataset
array must match the number and names of the columns in the Values
dataset array.

4-15

Microarray Analysis

The following illustrates a dataset array containing the measured value of

each variable per sample or feature:

A
B
C
D
E
F

Gender
'Male'
'Male'
'Male'
'Male'
'Male'
'Male'

Age
8
8
8
8
8
8

Type
'Wild
'Wild
'Wild
'Wild
'Wild
'Wild

type'
type'
type'
type'
type'
type'

Strain
'129S6/SvEvTac'
'129S6/SvEvTac'
'129S6/SvEvTac'
'A/J '
'A/J '
'C57BL/6J '

Source
'amygdala'
'amygdala'
'amygdala'
'amygdala'
'amygdala'
'amygdala'

The following illustrates a dataset array containing a list of the variable

names and their descriptions:

id
Gender
Age
Type
Strain
Source

VariableDescription
'Sample identifier'
'Gender of the mouse in study'
'The number of weeks since mouse birth'
'Genetic characters'
'The mouse strain'
'The tissue source for RNA collection'

A MetaData object lets you store, manage, and subset the metadata from a
microarray experiment. A MetaData object includes properties and methods
that let you access, retrieve, and change metadata from a microarray
experiment. These properties and methods are useful to view and analyze the
metadata. For a list of the properties and methods, see MetaData class

Constructing MetaData Objects

Constructing a MetaData Object from Two dataset Arrays
1 Import the bioma.data package so that the MetaData constructor function

is available.
import bioma.data.*
2 Load some sample data, which includes Fishers iris data of 5 measurements

on a sample of 150 irises.

4-16

Representing Sample and Feature Metadata in MetaData Objects

load fisheriris
3 Create a dataset array from some of Fishers iris data. The dataset

array will contain 750 measured values, one for each of 150 samples (iris
replicates) at five variables (species, SL, SW, PL, PW). In this dataset array,
the rows correspond to samples, and the columns correspond to variables.
irisValues = dataset({nominal(species),'species'}, ...
{meas, 'SL', 'SW', 'PL', 'PW'});
4 Create another dataset array containing a list of the variable names

and their descriptions. This dataset array will contain five rows, each
corresponding to the five variables: species, SL, SW, PL, and PW. The
first column will contain the variable name. The second column will have
a column header of VariableDescription and contain a description of
the variable.
% Create 5-by-1 cell array of description text for the variables
varDesc = {'Iris species', 'Sepal Length', 'Sepal Width', ...
'Petal Length', 'Petal Width'}';
% Create the dataset array from the variable descriptions
irisVarDesc = dataset(varDesc, ...
'ObsNames', {'species','SL','SW','PL','PW'}, ...
'VarNames', {'VariableDescription'})
irisVarDesc =

species
SL
SW
PL
PW

VariableDescription
'Iris species'
'Sepal Length'
'Sepal Width'
'Petal Length'
'Petal Width'

5 Create a MetaData object from the two dataset arrays.

MDObj1 = MetaData(irisValues, irisVarDesc);

4-17

Microarray Analysis

Constructing a MetaData Object from a Text File

1 Import the bioma.datapackage so that the MetaData constructor function

is available.
import bioma.data.*
2 View the mouseSampleData.txt file included with the Bioinformatics

Toolbox software.
Note that this text file contains two tables. One table contains 130
measured values, one for each of 26 samples (A through Z) at five variables
(Gender, Age, Type, Strain, and Source). In this table, the rows correspond
to samples, and the columns correspond to variables. The second table has
lines prefaced by the # symbol. It contains five rows, each corresponding to
the five variables: Gender, Age, Type, Strain, and Source. The first column
contains the variable name. The second column has a column header of
VariableDescription and contains a description of the variable.
# id: Sample identifier
# Gender: Gender of the mouse in study
# Age: The number of weeks since mouse birth
# Type: Genetic characters
# Strain: The mouse strain
# Source: The tissue source for RNA collection
ID Gender Age Type Strain Source
A Male 8 Wild type 129S6/SvEvTac amygdala
B Male 8 Wild type 129S6/SvEvTac amygdala
C Male 8 Wild type 129S6/SvEvTac amygdala
D Male 8 Wild type A/J amygdala
E Male 8 Wild type A/J amygdala
F Male 8 Wild type C57BL/6J amygdala
G Male 8 Wild type C57BL/6J amygdala
H Male 8 Wild type 129S6/SvEvTac cingulate cortex
I Male 8 Wild type 129S6/SvEvTac cingulate cortex
J Male 8 Wild type A/J cingulate cortex
K Male 8 Wild type A/J cingulate cortex
L Male 8 Wild type A/J cingulate cortex
M Male 8 Wild type C57BL/6J cingulate cortex
N Male 8 Wild type C57BL/6J cingulate cortex

4-18

Representing Sample and Feature Metadata in MetaData Objects

O
P
Q
R
S
T
U
V
W
X
Y
Z

Male
Male
Male
Male
Male
Male
Male
Male
Male
Male
Male
Male

8
8
8
8
8
8
8
8
8
8
8
8

Wild
Wild
Wild
Wild
Wild
Wild
Wild
Wild
Wild
Wild
Wild
Wild

type
type
type
type
type
type
type
type
type
type
type
type

129S6/SvEvTac hippocampus
129S6/SvEvTac hippocampus
A/J hippocampus
A/J hippocampus
C57BL/6J hippocampus
C57BL/6J4 hippocampus
129S6/SvEvTac hypothalamus
129S6/SvEvTac hypothalamus
A/J hypothalamus
A/J hypothalamus
C57BL/6J hypothalamus
C57BL/6J hypothalamus

3 Create a MetaData object from the metadata in the mouseSampleData.txt

file.
MDObj2 = MetaData('File', 'mouseSampleData.txt', 'VarDescChar', '#')
Sample Names:
A, B, ...,Z (26 total)
Variable Names and Meta Information:
VariableDescription
Gender

' Gender of the mouse in study'

Age

' The number of weeks since mouse birth'

Type

' Genetic characters'

Strain

' The mouse strain'

Source

' The tissue source for RNA collection'

For complete information on constructing MetaData objects, see MetaData

class.

Using Properties of a MetaData Object

To access properties of a MetaData object, use the following syntax:
objectname.propertyname

For example, to determine the number of variables in a MetaData object:

MDObj2.NVariables

4-19

Microarray Analysis

ans =
5

To set properties of a MetaData object, use the following syntax:

objectname.propertyname = propertyvalue

For example, to set the Description property of a MetaData object:

MDObj1.Description = 'This is my MetaData object for my sample metadata'

Note Property names are case sensitive. For a list and description of all
properties of a MetaData object, see MetaData class.

Using Methods of a MetaData Object

To use methods of a MetaData object, use either of the following syntaxes:
objectname.methodname

or
methodname(objectname)

For example, to access the dataset array in a MetaData object that contains
the variable values:
MDObj2.variableValues;

To access the dataset array of a MetaData object that contains the variable
descriptions:
variableDesc(MDObj2)
ans =

Gender
Age

4-20

VariableDescription
' Gender of the mouse in study'
' The number of weeks since mouse birth'

Representing Sample and Feature Metadata in MetaData Objects

Type
Strain
Source

' Genetic characters'

' The mouse strain'
' The tissue source for RNA collection'

Note For a complete list of methods of a MetaData object, see MetaData

class.

4-21

Microarray Analysis

Representing Experiment Information in a MIAME Object

In this section...
Overview of MIAME Objects on page 4-22
Constructing MIAME Objects on page 4-22
Using Properties of a MIAME Object on page 4-25
Using Methods of a MIAME Object on page 4-25

Overview of MIAME Objects

You can store information about experimental methods and conditions from
a microarray gene expression experiment in a MIAME object. It loosely
follows the Minimum Information About a Microarray Experiment (MIAME)
specification. It can include information about:
Experiment design
Microarrays used
Samples used
Sample preparation and labeling
Hybridization procedures and parameters
Normalization controls
Preprocessing information
Data processing specifications
A MIAME object includes properties and methods that let you access, retrieve,
and change experiment information related to a microarray experiment.
These properties and methods are useful to view and analyze the information.
For a list of the properties and methods, see MIAME class.

Constructing MIAME Objects

For complete information on constructing MIAME objects, see MIAME class.

4-22

Representing Experiment Information in a MIAME Object

Constructing a MIAME Object from a GEO Structure

1 Import the bioma.data package so that the MIAME constructor function

is available.
import bioma.data.*
2 Use the getgeodata function to return a MATLAB structure containing

Gene Expression Omnibus (GEO) Series data related to accession number

GSE4616.
geoStruct = getgeodata('GSE4616')
geoStruct =
Header: [1x1 struct]
Data: [12488x12 bioma.data.DataMatrix]
3 Use the MIAME constructor function to create a MIAME object from the

structure.
MIAMEObj1 = MIAME(geoStruct);
4 Display information about the MIAME object, MIAMEObj.
MIAMEObj1
MIAMEObj1 =
Experiment Description:
Author name: Mika,,Silvennoinen
Riikka,,Kivel
Maarit,,Lehti
Anna-Maria,,Touvras
Jyrki,,Komulainen
Veikko,,Vihko
Heikki,,Kainulainen
Laboratory: LIKES - Research Center
Contact information: Mika,,Silvennoinen
URL:
PubMedIDs: 17003243

4-23

Microarray Analysis

Abstract: A 90 word abstract is available. Use the Abstract property.

Experiment Design: A 234 word summary is available. Use the ExptDesign property.
Other notes:
[1x80 char]

Constructing a MIAME Object from Properties

1 Import the bioma.data package so that theMIAME constructor function

is available.
import bioma.data.*
2 Use the MIAME constructor function to create a MIAME object using

individual properties.
MIAMEObj2 = MIAME('investigator', 'Jane Researcher',...
'lab', 'One Bioinformatics Laboratory',...
'contact', 'jresearcher@lab.not.exist',...
'url', 'www.lab.not.exist',...
'title', 'Normal vs. Diseased Experiment',...
'abstract', 'Example of using expression data',...
'other', {'Notes:Created from a text file.'});

3 Display information about the MIAME object, MIAMEObj2.

MIAMEObj2
MIAMEObj2 =
Experiment Description:
Author name: Jane Researcher
Laboratory: One Bioinformatics Laboratory
Contact information: jresearcher@lab.not.exist
URL: www.lab.not.exist
PubMedIDs:
Abstract: A 4 word abstract is available. Use the Abstract property.
No experiment design summary available.
Other notes:
'Notes:Created from a text file.'

4-24

Representing Experiment Information in a MIAME Object

Using Properties of a MIAME Object

To access properties of a MIAME object, use the following syntax:
objectname.propertyname

For example, to retrieve the PubMed identifier of publications related to a

MIAME object:
MIAMEObj1.PubMedID
ans =
17003243

To set properties of a MIAME object, use the following syntax:

objectname.propertyname = propertyvalue

For example, to set the Laboratory property of a MIAME object:

MIAMEObj1.Laboratory = 'XYZ Lab'

Note Property names are case sensitive. For a list and description of all
properties of a MIAME object, see MIAME class.

Using Methods of a MIAME Object

To use methods of a MIAME object, use either of the following syntaxes:
objectname.methodname

or
methodname(objectname)

For example, to determine if a MIAME object is empty:

MIAMEObj1.isempty
ans =

4-25

Microarray Analysis

Note For a complete list of methods of a MIAME object, see MIAME class.

4-26

Representing All Data in an ExpressionSet Object

Representing All Data in an ExpressionSet Object

In this section...
Overview of ExpressionSet Objects on page 4-27
Constructing ExpressionSet Objects on page 4-29
Using Properties of an ExpressionSet Object on page 4-30
Using Methods of an ExpressionSet Object on page 4-30

Overview of ExpressionSet Objects

You can store all microarray experiment data and information in one object
by assembling the following into an ExpressionSet object:
One ExptData object containing expression values from a microarray
experiment in one or more DataMatrix objects
One MetaData object containing sample metadata in two dataset arrays
One MetaData object containing feature metadata in two dataset arrays
One MIAME object containing experiment descriptions
The following graphic illustrates a typical ExpressionSet object and its
component objects.

4-27

4-28

Microarray Analysis

Representing All Data in an ExpressionSet Object

Each element (DataMatrix object) in the ExpressionSet object has an element

name. Also, there is always one DataMatrix object whose element name is
Expressions.
An ExpressionSet object lets you store, manage, and subset the data from a
microarray gene expression experiment. An ExpressionSet object includes
properties and methods that let you access, retrieve, and change data,
metadata, and other information about the microarray experiment. These
properties and methods are useful to view and analyze the data. For a list of
the properties and methods, see ExpressionSet class.

Constructing ExpressionSet Objects

Note The following procedure assumes you have executed the example code
in the previous sections:
Representing Expression Data Values in ExptData Objects on page 4-11
Representing Sample and Feature Metadata in MetaData Objects on
page 4-15
Representing Experiment Information in a MIAME Object on page 4-22

1 Import the bioma package so that the ExpresssionSet constructor function

is available.
import bioma.*
2 Construct an ExpressionSet object from EDObj, an ExptData object, MDObj2,

a MetaData object containing sample variable information, and MIAMEObj,

a MIAME object.
ESObj = ExpressionSet(EDObj, 'SData', MDObj2, 'EInfo', MIAMEObj1);
3 Display information about the ExpressionSet object, ESObj.

ESObj

4-29

Microarray Analysis

ExpressionSet
Experiment Data: 500 features, 26 samples
Element names: Expressions
Sample Data:
Sample names:
A, B, ...,Z (26 total)
Sample variable names and meta information:
Gender: Gender of the mouse in study
Age: The number of weeks since mouse birth
Type: Genetic characters
Strain: The mouse strain
Source: The tissue source for RNA collection
Feature Data: none
Experiment Information: use 'exptInfo(obj)'

For complete information on constructing ExpressionSet objects, see

ExpressionSet class.

Using Properties of an ExpressionSet Object

To access properties of an ExpressionSet object, use the following syntax:
objectname.propertyname

For example, to determine the number of samples in an ExpressionSet object:

ESObj.NSamples
ans =
26

Note Property names are case sensitive. For a list and description of all
properties of an ExpressionSet object, see ExpressionSet class.

Using Methods of an ExpressionSet Object

To use methods of an ExpressionSet object, use either of the following
syntaxes:

4-30

Representing All Data in an ExpressionSet Object

objectname.methodname

or
methodname(objectname)

For example, to retrieve the sample variable names from an ExpressionSet

object:
ESObj.sampleVarNames
ans =
'Gender'

'Age'

'Type'

'Strain'

'Source'

To retrieve the experiment information contained in an ExpressionSet object:

exptInfo(ESObj)
ans =
Experiment description
Author name: Mika,,Silvennoinen
Riikka,,Kivel
Maarit,,Lehti
Anna-Maria,,Touvras
Jyrki,,Komulainen
Veikko,,Vihko
Heikki,,Kainulainen
Laboratory: XYZ Lab
Contact information: Mika,,Silvennoinen
URL:
PubMedIDs: 17003243
Abstract: A 90 word abstract is available Use the Abstract property.
Experiment Design: A 234 word summary is available Use the ExptDesign property.
Other notes:
[1x80 char]

4-31

Microarray Analysis

Note For a complete list of methods of an ExpressionSet object, see

ExpressionSet class.

4-32

Visualizing Microarray Images

Visualizing Microarray Images

In this section...
Overview of the Mouse Example on page 4-33
Exploring the Microarray Data Set on page 4-34
Spatial Images of Microarray Data on page 4-36
Statistics of the Microarrays on page 4-46
Scatter Plots of Microarray Data on page 4-48

Overview of the Mouse Example

This example looks at the various ways to visualize microarray data. The
data comes from a pharmacological model of Parkinsons disease (PD) using
a mouse brain. The microarray data for this example is from Brown, V.M.,
Ossadtchi, A., Khan, A.H., Yee, S., Lacan, G., Melega, W.P., Cherry, S.R.,
Leahy, R.M., and Smith, D.J.; "Multiplex three dimensional brain gene
expression mapping in a mouse model of Parkinsons disease"; Genome
Research 12(6): 868-884 (2002).
The microarray data used in this example is available in a Web supplement
to the paper by Brown et al. and in the file mouse_a1pd.gpr included with
the Bioinformatics Toolbox software.
http://labs.pharmacology.ucla.edu/smithlab/genome_multiplex/

The microarray data is also available on the Gene Expression Omnibus Web
site at
http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE30

The GenePix GPR-formatted file mouse_a1pd.gpr contains the data for one
of the microarrays used in the study. This is data from voxel A1 of the brain
of a mouse in which a pharmacological model of Parkinsons disease (PD)
was induced using methamphetamine. The voxel sample was labeled with
Cy3 (green) and the control, RNA from a total (not voxelated) normal mouse
brain, was labeled with Cy5 (red). GPR formatted files provide a large amount
of information about the array, including the mean, median, and standard

4-33

Microarray Analysis

deviation of the foreground and background intensities of each spot at the

635 nm wavelength (the red, Cy5 channel) and the 532 nm wavelength (the
green, Cy3 channel).

Exploring the Microarray Data Set

This procedure illustrates how to import data from the Web into the MATLAB
environment, using data from a study about gene expression in mouse brains
as an example. See Overview of the Mouse Example on page 4-33.
1 Read data from a file into a MATLAB structure. For example, in the

MATLAB Command Window, type

pd = gprread('mouse_a1pd.gpr')

Information about the structure displays in the MATLAB Command

Window:
pd =
Header:
Data:
Blocks:
Columns:
Rows:
Names:
IDs:
ColumnNames:
Indices:
Shape:

[1x1 struct]
[9504x38 double]
[9504x1 double]
[9504x1 double]
[9504x1 double]
{9504x1 cell}
{9504x1 cell}
{38x1 cell}
[132x72 double]
[1x1 struct]

2 Access the fields of a structure using StructureName.FieldName. For

example, you can access the field ColumnNames of the structure pd by typing
pd.ColumnNames

The column names are shown below.

ans =
'X'
'Y'
'Dia.'

4-34

Visualizing Microarray Images

'F635 Median'
'F635 Mean'
'F635 SD'
'B635 Median'
'B635 Mean'
'B635 SD'
'% > B635+1SD'
'% > B635+2SD'
'F635 % Sat.'
'F532 Median'
'F532 Mean'
'F532 SD'
'B532 Median'
'B532 Mean'
'B532 SD'
'% > B532+1SD'
'% > B532+2SD'
'F532 % Sat.'
'Ratio of Medians'
'Ratio of Means'
'Median of Ratios'
'Mean of Ratios'
'Ratios SD'
'Rgn Ratio'
'Rgn R'
'F Pixels'
'B Pixels'
'Sum of Medians'
'Sum of Means'
'Log Ratio'
'F635 Median - B635'
'F532 Median - B532'
'F635 Mean - B635'
'F532 Mean - B532'
'Flags'
3 Access the names of the genes. For example, to list the first 20 gene names,

type
pd.Names(1:20)

4-35

Microarray Analysis

A list of the first 20 gene names is displayed:

ans =
'AA467053'
'AA388323'
'AA387625'
'AA474342'
'Myo1b'
'AA473123'
'AA387579'
'AA387314'
'AA467571'
''
'Spop'
'AA547022'
'AI508784'
'AA413555'
'AA414733'
''
'Snta1'
'AI414419'
'W14393'
'W10596'

Spatial Images of Microarray Data

This procedure illustrates how to visualize microarray data by plotting image
maps. The function maimage can take a microarray data structure and create
a pseudocolor image of the data arranged in the same order as the spots on
the array. In other words, maimage plots a spatial plot of the microarray.
This procedure uses data from a study of gene expression in mouse brains.
For a list of field names in the MATLAB structure pd, see Exploring the
Microarray Data Set on page 4-34.
1 Plot the median values for the red channel. For example, to plot data from

the field F635 Median, type

figure
maimage(pd,'F635 Median')

4-36

Visualizing Microarray Images

The MATLAB software plots an image showing the median pixel values for
the foreground of the red (Cy5) channel.

2 Plot the median values for the green channel. For example, to plot data

from the field F532 Median, type

figure
maimage(pd,'F532 Median')

4-37

Microarray Analysis

The MATLAB software plots an image showing the median pixel values of
the foreground of the green (Cy3) channel.

3 Plot the median values for the red background. The field B635 Median

shows the median values for the background of the red channel.
figure
maimage(pd,'B635 Median')

4-38

Visualizing Microarray Images

The MATLAB software plots an image for the background of the red
channel. Notice the very high background levels down the right side of
the array.

4 Plot the medial values for the green background. The field B532 Median

shows the median values for the background of the green channel.
figure
maimage(pd,'B532 Median')

4-39

Microarray Analysis

The MATLAB software plots an image for the background of the green
channel.

5 The first array was for the Parkinsons disease model mouse. Now read in

the data for the same brain voxel but for the untreated control mouse. In
this case, the voxel sample was labeled with Cy3 and the control, total
brain (not voxelated), was labeled with Cy5.
wt = gprread('mouse_a1wt.gpr')

The MATLAB software creates a structure and displays information about

the structure.

4-40

Visualizing Microarray Images

wt =
Header:
Data:
Blocks:
Columns:
Rows:
Names:
IDs:
ColumnNames:
Indices:
Shape:

[1x1 struct]
[9504x38 double]
[9504x1 double]
[9504x1 double]
[9504x1 double]
{9504x1 cell}
{9504x1 cell}
{38x1 cell}
[132x72 double]
[1x1 struct]

6 Use the function maimage to show pseudocolor images of the foreground

and background. You can use the function subplot to put all the plots
onto one figure.
figure
subplot(2,2,1);
maimage(wt,'F635
subplot(2,2,2);
maimage(wt,'F532
subplot(2,2,3);
maimage(wt,'B635
subplot(2,2,4);
maimage(wt,'B532

Median')
Median')
Median')
Median')

4-41

Microarray Analysis

The MATLAB software plots the images.

7 If you look at the scale for the background images, you will notice that the

background levels are much higher than those for the PD mouse and there
appears to be something nonrandom affecting the background of the Cy3
channel of this slide. Changing the colormap can sometimes provide more
insight into what is going on in pseudocolor plots. For more control over the
color, try the colormapeditor function.
colormap hot

4-42

Visualizing Microarray Images

The MATLAB software plots the images.

8 The function maimage is a simple way to quickly create pseudocolor images

of microarray data. However if you want more control over plotting, it is

easy to create your own plots using the function imagesc.
First find the column number for the field of interest.
b532MedCol = find(strcmp(wt.ColumnNames,'B532 Median'))

The MATLAB software displays:

b532MedCol =
16
9 Extract that column from the field Data.

b532Data = wt.Data(:,b532MedCol);

4-43

Microarray Analysis

10 Use the field Indices to index into the Data.

figure
subplot(1,2,1);
imagesc(b532Data(wt.Indices))
axis image
colorbar
title('B532 Median')

The MATLAB software plots the image.

4-44

Visualizing Microarray Images

11 Bound the intensities of the background plot to give more contrast in the

image.
maskedData = b532Data;
maskedData(b532Data<500) = 500;
maskedData(b532Data>2000) = 2000;
subplot(1,2,2);
imagesc(maskedData(wt.Indices))
axis image
colorbar
title('Enhanced B532 Median')

The MATLAB software plots the images.

4-45

Microarray Analysis

Statistics of the Microarrays

This procedure illustrates how to visualize distributions in microarray data.
You can use the function maboxplot to look at the distribution of data in
each of the blocks.
1 In the MATLAB Command Window, type
figure
subplot(2,1,1)
maboxplot(pd,'F532 Median','title','Parkinson''s Disease Model Mouse')
subplot(2,1,2)
maboxplot(pd,'B532 Median','title','Parkinson''s Disease Model Mouse')
figure
subplot(2,1,1)
maboxplot(wt,'F532 Median','title','Untreated Mouse')
subplot(2,1,2)
maboxplot(wt,'B532 Median','title','Untreated Mouse')

The MATLAB software plots the images.

4-46

Visualizing Microarray Images

4-47

Microarray Analysis

2 Compare the plots.

From the box plots you can clearly see the spatial effects in the background
intensities. Blocks numbers 1, 3, 5, and 7 are on the left side of the
arrays, and numbers 2, 4, 6, and 8 are on the right side. The data must be
normalized to remove this spatial bias.

Scatter Plots of Microarray Data

This procedure illustrates how to visualize expression levels in microarray
data. There are two columns in the microarray data structure labeled 'F635
Median - B635' and 'F532 Median - B532'. These columns are the
differences between the median foreground and the median background for
the 635 nm channel and 532 nm channel respectively. These give a measure of
the actual expression levels, although since the data must first be normalized
to remove spatial bias in the background, you should be careful about using
these values without further normalization. However, in this example no
normalization is performed.

4-48

Visualizing Microarray Images

1 Rather than working with data in a larger structure, it is often easier to

extract the column numbers and data into separate variables.

cy5DataCol = find(strcmp(wt.ColumnNames,'F635 Median - B635'))
cy3DataCol = find(strcmp(wt.ColumnNames,'F532 Median - B532'))
cy5Data = pd.Data(:,cy5DataCol);
cy3Data = pd.Data(:,cy3DataCol);

The MATLAB software displays:

cy5DataCol =
34
cy3DataCol =
35
2 A simple way to compare the two channels is with a loglog plot. The

function maloglog is used to do this. Points that are above the diagonal in
this plot correspond to genes that have higher expression levels in the A1
voxel than in the brain as a whole.
figure
maloglog(cy5Data,cy3Data)
xlabel('F635 Median - B635 (Control)');
ylabel('F532 Median - B532 (Voxel A1)');

The MATLAB software displays the following messages and plots the
images.
Warning: Zero values are ignored
(Type "warning off Bioinfo:MaloglogZeroValues" to suppress
this warning.)
Warning: Negative values are ignored.
(Type "warning off Bioinfo:MaloglogNegativeValues" to suppress
this warning.)

4-49

Microarray Analysis

Notice that this function gives some warnings about negative and zero
elements. This is because some of the values in the 'F635 Median - B635'
and 'F532 Median - B532' columns are zero or even less than zero. Spots
where this happened might be bad spots or spots that failed to hybridize.
Points with positive, but very small, differences between foreground and
background should also be considered to be bad spots.
3 Disable the display of warnings by using the warning command. Although

warnings can be distracting, it is good practice to investigate why the

warnings occurred rather than simply to ignore them. There might be some
systematic reason why they are bad.
warnState = warning;

% First save the current warning

state.
% Now turn off the two warnings.
warning('off','Bioinfo:MaloglogZeroValues');
warning('off','Bioinfo:MaloglogNegativeValues');

4-50

Visualizing Microarray Images

figure
maloglog(cy5Data,cy3Data)
% Create the loglog plot
warning(warnState);
% Reset the warning state.
xlabel('F635 Median - B635 (Control)');
ylabel('F532 Median - B532 (Voxel A1)');

The MATLAB software plots the image.

4 An alternative to simply ignoring or disabling the warnings is to remove

the bad spots from the data set. You can do this by finding points where
either the red or green channel has values less than or equal to a threshold
value. For example, use a threshold value of 10.
threshold = 10;
badPoints = (cy5Data <= threshold) | (cy3Data <= threshold);

4-51

Microarray Analysis

The MATLAB software plots the image.

5 You can then remove these points and redraw the loglog plot.

cy5Data(badPoints) = []; cy3Data(badPoints) = [];

figure
maloglog(cy5Data,cy3Data)
xlabel('F635 Median - B635 (Control)');
ylabel('F532 Median - B532 (Voxel A1)');

4-52

Visualizing Microarray Images

The MATLAB software plots the image.

This plot shows the distribution of points but does not give any indication
about which genes correspond to which points.
6 Add gene labels to the plot. Because some of the data points have

been removed, the corresponding gene IDs must also be removed from
the data set before you can use them. The simplest way to do that is
wt.IDs(~badPoints).
maloglog(cy5Data,cy3Data,'labels',wt.IDs(~badPoints),...
'factorlines',2)
xlabel('F635 Median - B635 (Control)');
ylabel('F532 Median - B532 (Voxel A1)');

4-53

Microarray Analysis

The MATLAB software plots the image.

7 Try using the mouse to click some of the outlier points.

You will see the gene ID associated with the point. Most of the outliers are
below the y = x line. In fact, most of the points are below this line. Ideally
the points should be evenly distributed on either side of this line.
8 Normalize the points to evenly distribute them on either side of the line.

Use the function mameannorm to perform global mean normalization.

normcy5 = mameannorm(cy5Data);
normcy3 = mameannorm(cy3Data);

If you plot the normalized data you will see that the points are more evenly
distributed about the y = x line.
figure

4-54

Visualizing Microarray Images

maloglog(normcy5,normcy3,'labels',wt.IDs(~badPoints),...
'factorlines',2)
xlabel('F635 Median - B635 (Control)');
ylabel('F532 Median - B532 (Voxel A1)');

The MATLAB software plots the image.

9 The function mairplot is used to create an Intensity vs. Ratio plot for the

normalized data. This function works in the same way as the function
maloglog.
figure
mairplot(normcy5,normcy3,'labels',wt.IDs(~badPoints),...
'factorlines',2)

4-55

Microarray Analysis

The MATLAB software plots the image.

10 You can click the points in this plot to see the name of the gene associated

with the plot.

4-56

Analyzing Gene Expression Profiles

Analyzing Gene Expression Profiles

In this section...
Overview of the Yeast Example on page 4-57
Exploring the Data Set on page 4-57
Filtering Genes on page 4-61
Clustering Genes on page 4-64
Principal Component Analysis on page 4-68

Overview of the Yeast Example

This example demonstrates a number of ways to look for patterns in gene
expression profiles, using gene expression data from yeast shifting from
fermentation to respiration.
The microarray data for this example is from DeRisi, J.L., Iyer, V.R., and
Brown, P.O. (Oct 24, 1997). Exploring the metabolic and genetic control of
gene expression on a genomic scale. Science, 278 (5338), 680686. PMID:
9381177.
The authors used DNA microarrays to study temporal gene expression of
almost all genes in Saccharomyces cerevisiae during the metabolic shift from
fermentation to respiration. Expression levels were measured at seven time
points during the diauxic shift. The full data set can be downloaded from the
Gene Expression Omnibus Web site at:
http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE28

Exploring the Data Set

This procedure illustrates how to import data from the Web into the MATLAB
environment. The data for this procedure is available in the MAT-file
yeastdata.mat. This file contains the VALUE data or LOG_RAT2N_MEAN,
or log2 of ratio of CH2DN_MEAN and CH1DN_MEAN from the seven time
steps in the experiment, the names of the genes, and an array of the times at
which the expression levels were measured.
1 Load data into the MATLAB environment.

4-57

Microarray Analysis

load yeastdata.mat
2 Get the size of the data by typing

numel(genes)

The number of genes in the data set displays in the MATLAB Command
Window. The MATLAB variable genes is a cell array of the gene names.
ans =
6400
3 Access the entries using cell array indexing.

genes{15}

This displays the 15th row of the variable yeastvalues, which contains
expression levels for the open reading frame (ORF) YAL054C.
ans =
YAL054C
4 Use the function web to access information about this ORF in the

Saccharomyces Genome Database (SGD).

url = sprintf(...
'http://genome-www4.stanford.edu/cgi-bin/SGD/...
locus.pl?locus=%s',...
genes{15});
web(url);
5 A simple plot can be used to show the expression profile for this ORF.

plot(times, yeastvalues(15,:))
xlabel('Time (Hours)');
ylabel('Log2 Relative Expression Level');

4-58

Analyzing Gene Expression Profiles

The MATLAB software plots the figure. The values are log2 ratios.

6 Plot the actual values.

plot(times, 2.^yeastvalues(15,:))
xlabel('Time (Hours)');
ylabel('Relative Expression Level');

4-59

Microarray Analysis

The MATLAB software plots the figure. The gene associated with this
ORF, ACS1, appears to be strongly up-regulated during the diauxic shift.

7 Compare other genes by plotting multiple lines on the same figure.

hold on
plot(times, 2.^yeastvalues(16:26,:)')
xlabel('Time (Hours)');
ylabel('Relative Expression Level');
title('Profile Expression Levels');

4-60

Analyzing Gene Expression Profiles

The MATLAB software plots the image.

Filtering Genes
This procedure illustrates how to filter the data by removing genes that are
not expressed or do not change. The data set is quite large and a lot of the
information corresponds to genes that do not show any interesting changes
during the experiment. To make it easier to find the interesting genes, reduce
the size of the data set by removing genes with expression profiles that do not
show anything of interest. There are 6400 expression profiles. You can use
a number of techniques to reduce the number of expression profiles to some
subset that contains the most significant genes.
1 If you look through the gene list you will see several spots marked as

'EMPTY'. These are empty spots on the array, and while they might have
data associated with them, for the purposes of this example, you can

4-61

Microarray Analysis

consider these points to be noise. These points can be found using the
strcmp function and removed from the data set with indexing commands.
emptySpots = strcmp('EMPTY',genes);
yeastvalues(emptySpots,:) = [];
genes(emptySpots) = [];
numel(genes)

The MATLAB software displays:

ans =
6314

In the yeastvalues data you will also see several places where the
expression level is marked as NaN. This indicates that no data was collected
for this spot at the particular time step. One approach to dealing with
these missing values would be to impute them using the mean or median of
data for the particular gene over time. This example uses a less rigorous
approach of simply throwing away the data for any genes where one or
more expression levels were not measured.
2 Use the isnan function to identify the genes with missing data and then

use indexing commands to remove the genes.

nanIndices = any(isnan(yeastvalues),2);
yeastvalues(nanIndices,:) = [];
genes(nanIndices) = [];
numel(genes)

The MATLAB software displays:

ans =
6276

If you were to plot the expression profiles of all the remaining profiles,
you would see that most profiles are flat and not significantly different
from the others. This flat data is obviously of use as it indicates that the
genes associated with these profiles are not significantly affected by the
diauxic shift. However, in this example, you are interested in the genes
with large changes in expression accompanying the diauxic shift. You can
use filtering functions in the toolbox to remove genes with various types

4-62

Analyzing Gene Expression Profiles

of profiles that do not provide useful information about genes affected by

the metabolic change.
3 Use the function genevarfilter to filter out genes with small variance

over time. The function returns a logical array of the same size as the
variable genes with ones corresponding to rows of yeastvalues with
variance greater than the 10th percentile and zeros corresponding to those
below the threshold.
mask = genevarfilter(yeastvalues);
% Use the mask as an index into the values to remove the
% filtered genes.
yeastvalues = yeastvalues(mask,:);
genes = genes(mask);
numel(genes)

The MATLAB software displays:

ans =
5648
4 The function genelowvalfilter removes genes that have very low

absolute expression values. Note that the gene filter functions can also
automatically calculate the filtered data and names.
[mask, yeastvalues, genes] = genelowvalfilter(yeastvalues,genes,...
'absval',log2(4));
numel(genes)

The MATLAB software displays:

ans =
423
5 Use the function geneentropyfilter to remove genes whose profiles have

low entropy:
[mask, yeastvalues, genes] = geneentropyfilter(yeastvalues,genes,...
'prctile',15);
numel(genes)

4-63

Microarray Analysis

The MATLAB software displays:

ans =

310

Clustering Genes
Now that you have a manageable list of genes, you can look for relationships
between the profiles using some different clustering techniques from the
Statistics Toolbox software.
1 For hierarchical clustering, the function pdist calculates the pairwise

distances between profiles, and the function linkage creates the

hierarchical cluster tree.
corrDist = pdist(yeastvalues, 'corr');
clusterTree = linkage(corrDist, 'average');
2 The function cluster calculates the clusters based on either a cutoff

distance or a maximum number of clusters. In this case, the 'maxclust'

option is used to identify 16 distinct clusters.
clusters = cluster(clusterTree, 'maxclust', 16);
3 The profiles of the genes in these clusters can be plotted together using a

simple loop and the function subplot.

figure
for c = 1:16
subplot(4,4,c);
plot(times,yeastvalues((clusters == c),:)');
axis tight
end
suptitle('Hierarchical Clustering of Profiles');

The MATLAB software plots the images.

4-64

Analyzing Gene Expression Profiles

4 The Statistics Toolbox software also has a K-means clustering function.

Again, 16 clusters are found, but because the algorithm is different these
are not necessarily the same clusters as those found by hierarchical
clustering.
[cidx, ctrs] = kmeans(yeastvalues, 16,...
'dist','corr',...
'rep',5,...
'disp','final');
figure
for c = 1:16
subplot(4,4,c);
plot(times,yeastvalues((cidx == c),:)');
axis tight
end
suptitle('K-Means Clustering of Profiles');

4-65

Microarray Analysis

The MATLAB software displays:

13
14
26
22
26

iterations,
iterations,
iterations,
iterations,
iterations,

total
total
total
total
total

sum
sum
sum
sum
sum

of
of
of
of
of

distances
distances
distances
distances
distances

=
=
=
=
=

11.4042
8.62674
8.86066
9.77676
9.01035

5 Instead of plotting all of the profiles, you can plot just the centroids.

figure
for c = 1:16
subplot(4,4,c);
plot(times,ctrs(c,:)');
axis tight
axis off
% turn off the axis
end
suptitle('K-Means Clustering of Profiles');

4-66

Analyzing Gene Expression Profiles

The MATLAB software plots the figure:

6 You can use the function clustergram to create a heat map and

dendrogram from the output of the hierarchical clustering.

figure
clustergram(yeastvalues(:,2:end),'RowLabels',genes,...
'ColumnLabels',times(2:end))

4-67

Microarray Analysis

The MATLAB software plots the figure:

Principal Component Analysis

Principal-component analysis (PCA) is a useful technique you can use to
reduce the dimensionality of large data sets, such as those from microarray
analysis. You can also use PCA to find signals in noisy data.
1 Use the pca function in the Statistics Toolbox software to calculate the

principal components of a data set.

[pc, zscores, pcvars] = pca(yeastvalues)

The MATLAB software displays:

pc =
Columns 1 through 4

4-68

Analyzing Gene Expression Profiles

-0.0245
0.0186
0.0713
0.2254
0.2950
0.6596
0.6490

-0.3033
-0.5309
-0.1970
-0.2941
-0.6422
0.1788
0.2377

-0.1710
-0.3843
0.2493
0.1667
0.1415
0.5155
-0.6689

-0.2831
-0.5419
0.4042
0.1705
0.3358
-0.5032
0.2601

Columns 5 through 7
-0.1155
-0.2384
-0.7452
-0.2385
0.5592
-0.0194
-0.0673

0.4034
-0.2903
-0.3657
0.7520
-0.2110
-0.0961
-0.0039

0.7887
-0.3679
0.2035
-0.4283
0.1032
0.0667
0.0521

2 You can use the function cumsum to see the cumulative sum of the variances.

cumsum(pcvars./sum(pcvars) * 100)

The MATLAB software displays:

ans =
78.3719
89.2140
93.4357
96.0831
98.3283
99.3203
100.0000

This shows that almost 90% of the variance is accounted for by the first
two principal components.
3 A scatter plot of the scores of the first two principal components shows that

there are two distinct regions. This is not unexpected, because the filtering

4-69

Microarray Analysis

process removed many of the genes with low variance or low information.
These genes would have appeared in the middle of the scatter plot.
figure
scatter(zscores(:,1),zscores(:,2));
xlabel('First Principal Component');
ylabel('Second Principal Component');
title('Principal Component Scatter Plot');

The MATLAB software plots the figure:

4 The gname function from the Statistics Toolbox software can be used to

identify genes on a scatter plot. You can select as many points as you like
on the scatter plot.
gname(genes);

When you have finished selecting points, press Enter.

5 An alternative way to create a scatter plot is with the gscatter function

from the Statistics Toolbox software. gscatter creates a grouped scatter

4-70

Analyzing Gene Expression Profiles

plot where points from each group have a different color or marker. You
can use clusterdata, or any other clustering function, to group the points.
figure
pcclusters = clusterdata(zscores(:,1:2),6);
gscatter(zscores(:,1),zscores(:,2),pcclusters)
xlabel('First Principal Component');
ylabel('Second Principal Component');
title('Principal Component Scatter Plot with Colored Clusters');
gname(genes) % Press enter when you finish selecting genes.

The MATLAB software plots the figure:

4-71

Microarray Analysis

Detecting DNA Copy Number Alteration in Array-Based

CGH Data
This example shows how to detect DNA copy number alterations in
genome-wide array-based comparative genomic hybridization (CGH) data.
Introduction

Copy number changes or alterations is a form of genetic variation in the

human genome [1]. DNA copy number alterations (CNAs) have been linked to
the development and progression of cancer and many diseases.
DNA microarray based comparative genomic hybridization (CGH) is a
technique allows simultaneous monitoring of copy number of thousands of
genes throughout the genome [2,3]. In this technique, DNA fragments or
"clones" from a test sample and a reference sample differentially labeled with
dyes (typically, Cy3 and Cy5) are hybridized to mapped DNA microarrays and
imaged. Copy number alterations are related to the Cy3 and Cy5 fluorescence
intensity ratio of the targets hybridized to each probe on a microarray.
Clones with normalized test intensities significantly greater than reference
intensities indicate copy number gains in the test sample at those positions.
Similarly, significantly lower intensities in the test sample are signs of
copy number loss. BAC (bacterial artificial chromosome) clone based CGH
arrays have a resolution in the order of one million base pairs (1Mb) [3].
Oligonucleotide and cDNA arrays provide a higher resolution of 50-100kb [2].
Array CGH log2-based intensity ratios provide useful information about
genome-wide CNAs. In humans, the normal DNA copy number is two for all
the autosomes. In an ideal situation, the normal clones would correspond to
a log2 ratio of zero. The log2 intensity ratios of a single copy loss would be
-1, and a single copy gain would be 0.58. The goal is to effectively identify
locations of gains or losses of DNA copy number.
The data in this example is the Coriell cell line BAC array CGH data analyzed
by Snijders et al.(2001). The Coriell cell line data is widely regarded as a "gold
standard" data set. You can download this data of normalized log2-based
intensity ratios and the supplemental table of known karyotypes from
http://www.nature.com/ng/journal/v29/n3/suppinfo/ng754_S1.html. You will

4-72

Detecting DNA Copy Number Alteration in Array-Based CGH Data

compare these cytogenically mapped alterations with the locations of gains or

losses identified with various functions of MATLAB and its toolboxes.
For this example, the Coriell cell line data are provided in a MAT file. The
data file coriell_baccgh.mat contains coriell_data, a structure containing
of the normalized average of the log2-based test to reference intensity ratios
of 15 fibroblast cell lines and their genomic positions. The BAC targets are
ordered by genome position beginning at 1p and ending at Xq.
load coriell_baccgh
coriell_data

coriell_data =
Sample:
Chromosome:
GenomicPosition:
Log2Ratio:
FISHMap:

{1x15 cell}
[2285x1 int8]
[2285x1 int32]
[2285x15 double]
{2285x1 cell}

Visualizing the Genome Profile of the Array CGH Data Set

You can plot the genome wide log2-based test/reference intensity ratios of
DNA clones. In this example, you will display the log2 intensity ratios for cell
line GM03576 for chromosomes 1 through 23.
Find the sample index for the CM03576 cell line.
sample = find(strcmpi(coriell_data.Sample, 'GM03576'))

sample =
8

To label chromosomes and draw the chromosome borders, you need to find
the number of data points of in each chromosome.

4-73

Microarray Analysis

chr_nums = zeros(1, 23);

chr_data_len = zeros(1,23);
for c = 1:23
tmp = coriell_data.Chromosome == c;
chr_nums(c) = find(tmp, 1, 'last');
chr_data_len(c) = length(find(tmp));
end
% Draw a vertical bar at the end of a chromosome to indicate the border
x_vbar = repmat(chr_nums, 3, 1);
y_vbar = repmat([2;-2;NaN], 1, 23);

% Label the autosomes with their chromosome numbers, and the sex chromosome
% with X.
x_label = chr_nums - ceil(chr_data_len/2);
y_label = zeros(1, length(x_label)) - 1.6;
chr_labels=num2str((1:1:23)');
chr_labels = cellstr(chr_labels);
chr_labels{end} = 'X';
figure;hold on
h_ratio = plot(coriell_data.Log2Ratio(:,sample), '.');
h_vbar = line(x_vbar, y_vbar, 'color', [0.8 0.8 0.8]);
h_text = text(x_label, y_label, chr_labels,...
'fontsize', 8, 'HorizontalAlignment', 'center');
h_axis = get(h_ratio, 'parent');
set(h_axis, 'xtick', [], 'ygrid', 'on', 'box', 'on',...
'xlim', [0 chr_nums(23)], 'ylim', [-1.5 1.5])
title(coriell_data.Sample{sample})
xlabel({'', 'Chromosome'})
ylabel('Log2(T/R)')
hold off

In the plot, borders between chromosomes are indicated by grey vertical bars.
The plot indicates that the GM03576 cell line is trisomic for chromosomes
2 and 21 [3].

4-74

Detecting DNA Copy Number Alteration in Array-Based CGH Data

You can also plot the profile of each chromosome in a genome. In this
example, you will display the log2 intensity ratios for each chromosome in cell
line GM05296 individually.
sample = find(strcmpi(coriell_data.Sample, 'GM05296'));
figure;
for c = 1:23
idx = coriell_data.Chromosome == c;
chr_y = coriell_data.Log2Ratio(idx, sample);
subplot(5,5,c);
hp = plot(chr_y, '.');
line([0, chr_data_len(c)], [0,0], 'color', 'r');
h_axis = get(hp, 'Parent');
set(h_axis, 'xtick', [], 'Box', 'on',...
'xlim', [0 chr_data_len(c)], 'ylim', [-1.5 1.5])
xlabel(['chr ' chr_labels{c}], 'FontSize', 8)
end
suptitle('GM05296');

The plot indicates the GM05296 cell line has a partial trisomy at chromosome
10 and a partial monosomy at chromosome 11.
Observe that the gains and losses of copy number are discrete. These
alterations occur in contiguous regions of a chromosome that cover several
clones to entitle chromosome.
The array-based CGH data can be quite noisy. Therefore, accurate
identification of chromosome regions of equal copy number that accounts for
the noise in the data requires robust computational methods. In the rest of
this example, you will work with the data of chromosomes 9, 10 and 11 of
the GM05296 cell line.
Initialize a structure array for the data of these three chromosomes.
GM05296_Data = struct('Chromosome', {9 10 11},...
'GenomicPosition', {[], [], []},...

4-75

Microarray Analysis

'Log2Ratio', {[], [], []},...

'SmoothedRatio', {[], [], []},...
'DiffRatio', {[], [], []},...
'SegIndex', {[], [], []});
Filtering and Smoothing Data

A simple approach to perform high-level smoothing is to use a nonparametric

filter. The function mslowess implements a linear fit to samples within a
shifting window, is this example you use a SPAN of 15 samples.
for iloop
idx =
chr_x
chr_y

= 1:length(GM05296_Data)
coriell_data.Chromosome == GM05296_Data(iloop).Chromosome;
= coriell_data.GenomicPosition(idx);
= coriell_data.Log2Ratio(idx, sample);

% Remove NaN data points

idx = ~isnan(chr_y);
GM05296_Data(iloop).GenomicPosition = double(chr_x(idx));
GM05296_Data(iloop).Log2Ratio = chr_y(idx);
% Smoother
GM05296_Data(iloop).SmoothedRatio = ...
mslowess(GM05296_Data(iloop).GenomicPosition,...
GM05296_Data(iloop).Log2Ratio,...
'SPAN',15);
% Find the derivative of the smoothed ratio
GM05296_Data(iloop).DiffRatio = ...
diff([0; GM05296_Data(iloop).SmoothedRatio]);
end

To better visualize and later validate the locations of copy number changes,
we need cytoband information. Read the human cytoband information from
the hs_cytoBand.txt data file using the cytobandread function. It returns a
structure of human cytoband information [4].
hs_cytobands = cytobandread('hs_cytoBand.txt')
% Find the centromere positions for the chromosomes.

4-76

Detecting DNA Copy Number Alteration in Array-Based CGH Data

acen_idx = strcmpi(hs_cytobands.GieStains, 'acen');

acen_ends = hs_cytobands.BandEndBPs(acen_idx);
% Convert the cytoband data from bp to kilo bp because the genomic
% positions in Coriell Cell Line data set are in kilo base pairs.
acen_pos = acen_ends(1:2:end)/1000;

hs_cytobands =
ChromLabels:
BandStartBPs:
BandEndBPs:
BandLabels:
GieStains:

{862x1
[862x1
[862x1
{862x1
{862x1

cell}
int32]
int32]
cell}
cell}

You can inspect the data by plotting the log2-based ratios, the smoothed ratios
and the derivative of the smoothed ratios together. You can also display the
centromere position of a chromosome in the data plots. The magenta vertical
bar marks the centromere of the chromosome.
for iloop = 1:length(GM05296_Data)
chr = GM05296_Data(iloop).Chromosome;
chr_x = GM05296_Data(iloop).GenomicPosition;
figure; hold on
plot(chr_x, GM05296_Data(iloop).Log2Ratio, '.');
line(chr_x, GM05296_Data(iloop).SmoothedRatio,...
'Color', 'r', 'LineWidth', 2);
line(chr_x, GM05296_Data(iloop).DiffRatio,...
'Color', 'k', 'LineWidth', 2);
line([acen_pos(chr), acen_pos(chr)], [-1, 1],...
'Color', 'm', 'LineWidth', 2, 'LineStyle', '-.');
if iloop == 1
legend('Raw','Smoothed','Diff', 'Centromere');
end
ylim([-1, 1])
xlabel('Genomic Position')
ylabel('Log2(T/R)')
title(sprintf('GM05296: Chromosome %d ', chr))

4-77

Microarray Analysis

hold off
end

Detecting Change-Points

The derivatives of the smoothed ratio over a certain threshold usually

indicate substantial changes with large peaks, and provide the estimate of the
change-point indices. For this example you will select a threshold of 0.1.
thrd = 0.1;
for iloop = 1:length(GM05296_Data)
idx = find(abs(GM05296_Data(iloop).DiffRatio) > thrd );
N = numel(GM05296_Data(iloop).SmoothedRatio);
GM05296_Data(iloop).SegIndex = [1;idx;N];
% Number of possible segments found
fprintf('%d segments initially found on Chromosome %d.\n',...
numel(GM05296_Data(iloop).SegIndex) - 1,...
GM05296_Data(iloop).Chromosome)
end
1 segments initially found on Chromosome 9.
4 segments initially found on Chromosome 10.
5 segments initially found on Chromosome 11.
Optimizing Change-Points by GM Clustering

Gaussian Mixture (GM) or Expectation-Maximization (EM) clustering can

provide fine adjustments to the change-point indices [5]. The convergence to
statistically optimal change-point indices can be facilitated by surrounding
each index with equal-length set of adjacent indices. Thus each edge is
associated with left and right distributions. The GM clustering learns the

4-78

Detecting DNA Copy Number Alteration in Array-Based CGH Data

maximum-likelihood parameters of the two distributions. It then optimally

adjusts the indices given the learned parameters.
You can set the length for the set of adjacent positions distributed around the
change-point indices. For this example, you will select a length of 5. You can
also inspect each change-point by plotting its GM clusters. In this example,
you will plot the GM clusters for the Chromosome 10 data.
len = 5;
for iloop = 1:length(GM05296_Data)
seg_num = numel(GM05296_Data(iloop).SegIndex) - 1;
if seg_num > 1
% Plot the data points in chromosome 10 data
if GM05296_Data(iloop).Chromosome == 10
figure; hold on;
plot(GM05296_Data(iloop).GenomicPosition,...
GM05296_Data(iloop).Log2Ratio, '.')
ylim([-0.5, 1])
xlabel('Genomic Position')
ylabel('Log2(T/R)')
title(sprintf('Chromosome %d - GM05296', ...
GM05296_Data(iloop).Chromosome))
end
segidx = GM05296_Data(iloop).SegIndex;
segidx_emadj = GM05296_Data(iloop).SegIndex;
for jloop = 2:seg_num
ileft = min(segidx(jloop) - len, segidx(jloop));
iright = max(segidx(jloop) + len, segidx(jloop));
gmx = GM05296_Data(iloop).GenomicPosition(ileft:iright);
gmy = GM05296_Data(iloop).SmoothedRatio(ileft:iright);

% Select initial guess for the of cluster index for each point.
gmpart = (gmy > (min(gmy) + range(gmy)/2)) + 1;
% Create a Gaussian mixture model object
gm = gmdistribution.fit(gmy, 2, 'start', gmpart);
gmid = gm.cluster(gmy);

4-79

Microarray Analysis

segidx_emadj(jloop) = find(abs(diff(gmid))==1) + ileft;

% Plot GM clusters for the change-points in chromosome 10 data
if GM05296_Data(iloop).Chromosome == 10
plot(gmx(gmid==1),gmy(gmid==1), 'g.',...
gmx(gmid==2), gmy(gmid==2), 'r.')
end
end
% Remove repeat indices
zeroidx = [diff(segidx_emadj) == 0; 0];
GM05296_Data(iloop).SegIndex = segidx_emadj(~zeroidx);
end

% Number of possible segments found

fprintf('%d segments found on Chromosome %d after GM clustering adjustm
numel(GM05296_Data(iloop).SegIndex) - 1,...
GM05296_Data(iloop).Chromosome)
end
hold off;
1 segments found on Chromosome 9 after GM clustering adjustment.
3 segments found on Chromosome 10 after GM clustering adjustment.
5 segments found on Chromosome 11 after GM clustering adjustment.

Testing Change-Point Significance

Once you determine the optimal change-point indices, you also need to
determine if each segment represents a statistically significant changes
in DNA copy number. You will perform permutation t-tests to assess the
significance of the segments identified. A segment includes all the data points
from one change-point to the next change-point or the chromosome end. In
this example, you will perform 10,000 permutations of the data points on two
consecutive segments along the chromosome at the significance level of 0.01.
alpha = 0.01;
for iloop = 1:length(GM05296_Data)
seg_num = numel(GM05296_Data(iloop).SegIndex) - 1;

4-80

Detecting DNA Copy Number Alteration in Array-Based CGH Data

seg_index = GM05296_Data(iloop).SegIndex;
if seg_num > 1
ppvals = zeros(seg_num+1, 1);
for sloop = 1:seg_num-1
seg1idx = seg_index(sloop):seg_index(sloop+1)-1;
if sloop== seg_num-1
seg2idx = seg_index(sloop+1):(seg_index(sloop+2));
else
seg2idx = seg_index(sloop+1):(seg_index(sloop+2)-1);
end
seg1 = GM05296_Data(iloop).SmoothedRatio(seg1idx);
seg2 = GM05296_Data(iloop).SmoothedRatio(seg2idx);
n1 = numel(seg1);
n2 = numel(seg2);
N = n1+n2;
segs = [seg1;seg2];
% Compute observed t statistics
t_obs = mean(seg1) - mean(seg2);

% Permutation test
iter = 10000;
t_perm = zeros(iter,1);
for i = 1:iter
randseg = segs(randperm(N));
t_perm(i) = abs(mean(randseg(1:n1))-mean(randseg(n1+1:N)));
end
ppvals(sloop+1) = sum(t_perm >= abs(t_obs))/iter;
end
sigidx = ppvals < alpha;
GM05296_Data(iloop).SegIndex = seg_index(sigidx);
end

% Number segments after significance tests

fprintf('%d segments found on Chromosome %d after significance tests.\n

4-81

Microarray Analysis

numel(GM05296_Data(iloop).SegIndex) - 1, GM05296_Data(iloop).Chromos
end
1 segments found on Chromosome 9 after significance tests.
3 segments found on Chromosome 10 after significance tests.
4 segments found on Chromosome 11 after significance tests.
Assessing Copy Number Alterations

Cytogenetic study indicates cell line GM05296 has a trisomy at 10q21-10q24

and a monosomy at 11p12-11p13 [3]. Plot the segment means of the three
chromosomes over the original data with bold red lines, and add the
chromosome ideograms to the plots using the chromosomeplot function. Note
that the genomic positions in the Coriell cell line data set are in kilo base
pairs. Therefore, you will need to convert cytoband data from bp to kilo bp
when adding the ideograms to the plot.

for iloop = 1:length(GM05296_Data)

figure;
seg_num = numel(GM05296_Data(iloop).SegIndex) - 1;
seg_mean = ones(seg_num,1);
chr_num = GM05296_Data(iloop).Chromosome;
for jloop = 2:seg_num+1
idx = GM05296_Data(iloop).SegIndex(jloop-1):GM05296_Data(iloop).Seg
seg_mean(idx) = mean(GM05296_Data(iloop).Log2Ratio(idx));
line(GM05296_Data(iloop).GenomicPosition(idx), seg_mean(idx),...
'color', 'r', 'linewidth', 3);
end
line(GM05296_Data(iloop).GenomicPosition, GM05296_Data(iloop).Log2Ratio
'linestyle', 'none', 'Marker', '.');
line([acen_pos(chr_num), acen_pos(chr_num)], [-1, 1],...
'linewidth', 2,...
'color', 'm',...
'linestyle', '-.');
ylabel('Log2(T/R)')
set(gca, 'Box', 'on', 'ylim', [-1, 1])
title(sprintf('Chromosome %d - GM05296', chr_num));
chromosomeplot(hs_cytobands, chr_num, 'addtoplot', gca,

4-82

'unit', 2)

Detecting DNA Copy Number Alteration in Array-Based CGH Data

end

As shown in the plots, no copy number alterations were found on chromosome

9, there is copy number gain span from 10q21 to 10q24, and a copy number
loss region from 11p12 to 11p13. The CNAs found match the known results in
cell line GM05296 determined by cytogenetic analysis.
You can also display the CNAs of the GM05296 cell line align to the
chromosome ideogram summary view using the chromosomeplot function.
Determine the genomic positions for the CNAs on chromosomes 10 and 11.
chr10_idx = GM05296_Data(2).SegIndex(2):GM05296_Data(2).SegIndex(3)-1;
chr10_cna_start = GM05296_Data(2).GenomicPosition(chr10_idx(1))*1000;
chr10_cna_end
= GM05296_Data(2).GenomicPosition(chr10_idx(end))*1000;
chr11_idx = GM05296_Data(3).SegIndex(2):GM05296_Data(3).SegIndex(3)-1;
chr11_cna_start = GM05296_Data(3).GenomicPosition(chr11_idx(1))*1000;
chr11_cna_end = GM05296_Data(3).GenomicPosition(chr11_idx(end))*1000;

Create a structure containing the copy number alteration data from

the GM05296 cell line data according to the input requirements of the
chromosomeplot function.
cna_struct = struct('Chromosome', [10 11],...
'CNVType', [2 1],...
'Start', [chr10_cna_start, chr11_cna_start],...
'End',
[chr10_cna_end, chr11_cna_end])

cna_struct =
Chromosome: [10 11]
CNVType: [2 1]
Start: [69209000 34420000]

4-83

Microarray Analysis

End: [105905000 35914000]

chromosomeplot(hs_cytobands, 'cnv', cna_struct, 'unit', 2)

title('Human Karyogram with Copy Number Alterations of GM05296')

This example shows how MATLAB and its toolboxes provide tools for the
analysis and visualization of copy-number alterations in array-based CGH
data.
References

[1] Redon, R., Ishikawa, S., Fitch, K.R., et al. (2006). Global variation in copy
number in the human genome. Nature 444, 444-454.
[2] Pinkel, D., Segraves, R., Sudar, D., Clark, S., Poole, I., Kowbel, D., Collins,
C. Kuo, W.L., Chen, C., Zhai, Y., et al. (1998). High resolution analysis of
DNA copy number variations using comparative genomic hybridization to
microarrays. Nat. Genet. 20, 207-211.
[3] Snijders, A.M., Nowak, N., Segraves, R., Blackwood, S., Brown, N., Conroy,
J., Hamilton, G., Hindle, A.K., Huey, B., Kimura, K., et al. (2001). Assembly
of microarrays for genome-wide measurement of DNA copy number", Nat.
Genet. 29, 263-264.
[4] Human Genome NCBI Build 36.
[5] Myers, C.L., Dunham, M.J., Kung, S.Y., and Troyanskaya, O.G. (2004).
Accurate detection of aneuploidies in array CGH and gene expression
microarray data. Bioinformatics 20, 18, 3533-3543.
Suggest an enhancement for this example.

4-84

Exploring Gene Expression Data

Exploring Gene Expression Data

This example shows how to identify differentially expressed genes. Then it
uses Gene Ontology to determine significant biological functions that are
associated to the down- and up-regulated genes.
Introduction

Microarrays contain oligonucleotide or cDNA probes for comparing the

expression profile of genes on a genomic scale. Determining if changes in
gene expression are statistically significant between different conditions,
e.g. two different tumor types, and determining the biological function
of the differentially expressed genes, are important aims in a microarray
experiment.
A publicly available dataset containing gene expression data of 42 tumor
tissues of the embryonal central nervous system (CNS, Pomeroy et al. 2002)
is used for this example. The samples were hybridized on Affymetrix
HuGeneFL GeneChip arrays.
The CNS dataset (CEL files) is available at the CNS experiment web site.
The 42 tumor tissue samples include 10 medulloblastomas, 10 rhabdoid, 10
malignant glioma, 8 supratentorial PNETS, and 4 normal human cerebella.
The CNS raw dataset was preprocessed with the Robust Multi-array Average
(RMA) and GC Robust Multi-array Average (GCRMA) procedures. For
further information on Affymetrix oligonucleotide microarray preprocessing,
see Preprocessing Affymetrix Microarray Data at the Probe Level.
You will use the t-test and false discovery rate to detect differentially
expressed genes between two of the tumor types. Additionally, you will look
at Gene Ontology terms related to the significantly up-regulated genes.
Loading the Expression Data

Load the MAT file cnsexpressiondata containing three DataMatrix

objects. Gene expression values preprocessed by RMA and GCRMA (MLE
and EB) procedures are stored in the DataMatrix objects expr_cns_rma,
expr_cns_gcrma_mle, and expr_cns_gcrma_eb respectively.
load cnsexpressiondata

4-85

Microarray Analysis

In each DataMatrix object, each row corresponds to a probe set on the

HuGeneFl array, and each column corresponds to a sample. The row
names are the probe set IDs and column names are the sample names. The
DataMatrix object expr_cns_gcrma_eb will be used in this example. You can
use either of the other two expression variables as well.
You can get the properties of the DataMatrix object expr_cns_gcrma_eb
using the get command.
get(expr_cns_gcrma_eb)
Name:
RowNames:
ColNames:
NRows:
NCols:
NDims:
ElementClass:

''
{7129x1 cell}
{1x42 cell}
7129
42
2
'single'

Determine the number of genes and number of samples in the DataMatrix

object expr_cns_gcrma_eb.
[nGenes, nSamples] = size(expr_cns_gcrma_eb)

nGenes =
7129

nSamples =
42

You can use gene symbols instead of the probe set IDs to label the expression
values. The gene symbols for the HuGeneFl array are provided in a MAT
file containing a Map object.

4-86

Exploring Gene Expression Data

load HuGeneFL_GeneSymbol_Map;
Warning: Unknown parameter name
of known parameters.

will be ignored.

Check documentation for

Create a cell array of gene symbols for the expression values from the
hu6800GeneSymbolMap object.
huGenes = values(hu6800GeneSymbolMap, expr_cns_gcrma_eb.RowNames);

Set the row names of the exprs_cns_gcrma_eb to gene symbols using the
rownames method of the DataMatrix object.
expr_cns_gcrma_eb = rownames(expr_cns_gcrma_eb, ':', huGenes);
Filtering the Expression Data

Remove gene expression data with empty gene symbols. In the example, the
empty symbols are labeled as '---'.
expr_cns_gcrma_eb('---', :) = [];

Many of the genes in this study are not expressed, or have only small
variability across the samples. Remove these genes using non-specific
filtering.
Use genelowvalfilter to filter out genes with very low absolute expression
values.
[mask, expr_cns_gcrma_eb] = genelowvalfilter(expr_cns_gcrma_eb);

Use genevarfilter to filter out genes with a small variance across samples.
[mask, expr_cns_gcrma_eb] = genevarfilter(expr_cns_gcrma_eb);

Determine the number of genes after filtering.

nGenes = expr_cns_gcrma_eb.NRows

nGenes =

4-87

Microarray Analysis

5669

Identifying Differential Gene Expression

You can now compare the gene expression values between two groups of data:
CNS medulloblastomas (MD) and non-neuronal origin malignant gliomas
(Mglio) tumor.
From the expression data of all 42 samples, extract the data of the 10 MD
samples and the 10 Mglio samples.
MDs = strncmp(expr_cns_gcrma_eb.ColNames,'Brain_MD', 8);
Mglios = strncmp(expr_cns_gcrma_eb.ColNames,'Brain_MGlio', 11);
MDData = expr_cns_gcrma_eb(:, MDs);
get(MDData)
Name:
RowNames:
ColNames:
NRows:
NCols:
NDims:
ElementClass:

''
{5669x1 cell}
{1x10 cell}
5669
10
2
'single'

MglioData = expr_cns_gcrma_eb(:, Mglios);

get(MglioData)
Name:
RowNames:
ColNames:
NRows:
NCols:
NDims:
ElementClass:

4-88

''
{5669x1 cell}
{1x10 cell}
5669
10
2
'single'

Exploring Gene Expression Data

A standard statistical test for detecting significant changes between the

measurement of a variable in two groups is the t-test. Conduct a t-test for
each gene to identify significant changes in expression values between the
MD samples and Mglio samples. You can inspect the test results from the
normal quantile plot of t-scores and the histograms of t-scores and p-values
of the t-tests.
[pvalues, tscores] = mattest(MDData, MglioData,...
'Showhist', true', 'Showplot', true);

In any test situation, two types of errors can occur, a false positive by
declaring that a gene is differentially expressed when it is not, and a false
negative when the test fails to identify a truly differentially expressed gene.
In multiple hypothesis testing, which simultaneously tests the null hypothesis
of thousands of genes using microarray expression data, each test has a
specific false positive rate, or a false discovery rate (FDR). False discovery
rate is defined as the expected ratio of the number of false positives to the
total number of positive calls in a differential expression analysis between
two groups of samples (Storey et al., 2003).
In this example, you will compute the FDR using the Storey-Tibshirani
procedure (Storey et al., 2003). The procedure also computes the q-value of
a test, which measures the minimum FDR that occurs when calling the test
significant. The estimation of FDR depends on the truly null distribution of
the multiple tests, which is unknown. Permutation methods can be used to
estimate the truly null distribution of the test statistics by permuting the
columns of the gene expression data matrix (Storey et al., 2003, Dudoit et
al., 2003). Depending on the sample size, it may not be feasible to consider
all possible permutations. Usually a random subset of permutations are
considered in the case of large sample size. Use the nchoosek function in
Statistics Toolbox to find out the number of all possible permutations of
the samples in this example.

all_possible_perms = nchoosek(1:MDData.NCols+MglioData.NCols, MDData.NCols)

size(all_possible_perms, 1)

4-89

Microarray Analysis

ans =
184756

Perform a permutation t-test using mattest and the PERMUTE option to

compute the p-values of 10,000 permutations by permuting the columns of the
gene expression data matrix of MDData and MglioData (Dudoit et al., 2003).
pvaluesCorr = mattest(MDData, MglioData, 'Permute', 10000);

Determine the number of genes considered to have statistical significance at

the p-value cutoff of 0.05. Note: You may get a different number of genes
due to the permutation test outcome.
cutoff = 0.05;
sum(pvaluesCorr < cutoff)

ans =
2121

Estimate the FDR and q-values for each test using mafdr. The quantity pi0 is
the overall proportion of true null hypotheses in the study. It is estimated
from the simulated null distribution via bootstrap or the cubic polynomial fit.
Note: You can also manually set the value of lambda for estimating pi0.
figure;
[pFDR, qvalues] = mafdr(pvaluesCorr, 'showplot', true);

Determine the number of genes that have q-values less than the cutoff value.
Note: You may get a different number of genes due to the permutation test
and the bootstrap outcomes.
sum(qvalues < cutoff)

4-90

Exploring Gene Expression Data

ans =
2173

Many genes with low FDR implies that the two groups, MD and Mglio, are
biologically distinct.
You can also empirically estimate the FDR adjusted p-values using the
Benjamini-Hochberg (BH) procedure (Benjamini et al, 1995) by setting the
mafdr input parameter BHFDR to true.
pvaluesBH = mafdr(pvaluesCorr, 'BHFDR', true);
sum(pvaluesBH < cutoff)

ans =
1374

You can store the t-scores, p-values, pFDRs, q-values and BH FDR corrected
p-values together as a DataMatrix object.
testResults = [tscores pvaluesCorr pFDR qvalues pvaluesBH];

Update the column name for BH FDR corrected p-values using the colnames
method of DataMatrix object.
testResults = colnames(testResults, 5, {'FDR_BH'});

You can sort by p-values pvaluesCorr using the sortrows mathod.

testResults = sortrows(testResults, 2);

Display the first 23 genes in testResults. Note: Your results may be

different from those shown below due to the permutation test and the
bootstrap outcomes.
testResults(1:23, :)

4-91

Microarray Analysis

ans =

PLEC1
HNRPA1
FCGR2A
PLEC1
FBL
KIAA0367
ID2B
RBMX
PAFAH1B3
H3F3A
LRP1
PEA15
ID2B
SFRS3
HLA-DPA1
C5orf13
PTMA
NAP1L1
HMGB2
RAB31
ARAF
PTPRZ1
SPARCL1

t-scores
-9.6223
9.359
-9.3548
-9.3495
9.1518
-8.996
-8.9285
8.8905
8.7561
8.6512
-8.6465
-8.3256
-8.1183
8.1166
-7.8546
7.7195
7.7013
7.674
7.6532
-13.664
-7.5549
-7.5352
-7.3639

p-values
6.7194e-09
1.382e-08
1.394e-08
1.4094e-08
1.9875e-08
2.4324e-08
2.6667e-08
2.8195e-08
3.5317e-08
4.5191e-08
4.6243e-08
1.1419e-07
1.7041e-07
1.7055e-07
2.4004e-07
2.9229e-07
2.9658e-07
3.0477e-07
3.123e-07
3.308e-07
4.7835e-07
4.9875e-07
7.8426e-07

FDR
1.3675e-05
1.4063e-05
9.457e-06
7.171e-06
8.0899e-06
8.2509e-06
7.7533e-06
7.1728e-06
7.9864e-06
9.1973e-06
8.5559e-06
1.9367e-05
2.6679e-05
2.4793e-05
3.2569e-05
3.7179e-05
3.5506e-05
3.446e-05
3.3452e-05
3.3662e-05
4.6359e-05
4.614e-05
6.9397e-05

q-values
7.171e-06
7.171e-06
7.171e-06
7.171e-06
7.1728e-06
7.1728e-06
7.1728e-06
7.1728e-06
7.9864e-06
8.5559e-06
8.5559e-06
1.9367e-05
2.4793e-05
2.4793e-05
3.2569e-05
3.3452e-05
3.3452e-05
3.3452e-05
3.3452e-05
3.3662e-05
4.614e-05
4.614e-05
6.2018e-05

A gene is considered to be differentially expressed between the two groups

of samples if it shows both statistical and biological significance. In this
example, you will compare the gene expression ratio of MD over Mglio
tumor samples. Therefore an up-regulated gene in this example has higher
expression in MD and down-regulate gene has higher expression in Mglio.
Plot the -log10 of p-values against the biological effect in a volcano plot. Note:
From the volcano plot UI, you can interactively change the p-value cutoff and
fold change limit, and export differentially expressed genes.

4-92

FDR_B
1.997
1.997
1.997
1.997
1.99
1.99
1.99
1.99
2.224
2.383
2.383
5.394
6.905
6.905
9.07
9.317
9.317
9.317
9.317
9.376
0.000
0.000
0.000

Exploring Gene Expression Data

diffStruct = mavolcanoplot(MDData, MglioData, pvaluesCorr)

diffStruct =
Name:
PVCutoff:
FCThreshold:
GeneLabels:
PValues:
FoldChanges:

'Differentially Expressed'
0.0500
2
{327x1 cell}
[327x1 bioma.data.DataMatrix]
[327x1 bioma.data.DataMatrix]

Ctrl-click genes in the gene lists to label the genes in the plot. As seen in the
volcano plot, genes specific for neuronal based cerebella granule cells, such
as ZIC and NEUROD, are found in the up-regulated gene list, while genes
typical of the astrocytic and oligodendrocytic lineage and cell differentiation,
such as SOX2, PEA15, and ID2B, are found in the down-regulated list.
Determine the number of differentially expressed genes.
nDiffGenes = diffStruct.PValues.NRows

nDiffGenes =
327

Get the list of up-regulated genes for MD compared to Mglio.

up_geneidx = find(diffStruct.FoldChanges > 0);
up_genes = rownames(diffStruct.FoldChanges, up_geneidx);
nUpGenes = length(up_geneidx)

nUpGenes =

4-93

Microarray Analysis

225

Determine the number of down-regulated genes for MD compared to Mglio.

nDownGenes = sum(diffStruct.FoldChanges < 0)

nDownGenes =
102

Annotating Up-Regulated Genes Using Gene Ontology

Use Gene Ontology (GO) to annotate the differentially expressed genes. You
can look at the up-regulated genes from the analysis above. Download the
Homo sapiens annotations (gene_association.goa_human.gz file) from
Gene Ontology Current Annotations, unzip, and store it in your the current
directory.
Find the indices of the up-regulated genes for Gene Ontology analysis.

huGenes = rownames(expr_cns_gcrma_eb);
for i = 1:nUpGenes
up_geneidx(i) = find(strncmpi(huGenes, up_genes{i}, length(up_genes{i})
end

Load the Gene Ontology database into a MATLAB object using the geneont
function.
GO = geneont('live',true);

Read the Homo sapiens gene annotation file. For this example, you will look
only at genes that are related to molecular function, so you only need to read
the information where the Aspect field is set to F. The fields that are of
interest are the gene symbol and associated ID. In GO Annotation files these
have field names DB_Object_Symbol and GOid respectively.
HGann = goannotread('gene_association.goa_human',...
'Aspect','F','Fields',{'DB_Object_Symbol','GOid'});

4-94

Exploring Gene Expression Data

Create a map between annotated genes and GO terms.

HGmap = containers.Map();
for i=1:numel(HGann)
key = HGann(i).DB_Object_Symbol;
if isKey(HGmap,key)
HGmap(key) = [HGmap(key) HGann(i).GOid];
else
HGmap(key) = HGann(i).GOid;
end
end

Determine the number of Homo sapiens annotated genes related to molecular

function.
HGmap.Count

ans =
16006

Not all of the 5758 genes on the HuGeneFL chip are annotated. For every
gene on the chip, see if it is annotated by comparing its gene symbol to the
list of gene symbols from GO. Track the number of annotated genes and the
number of up-regulated genes associated with each GO term. Note that data
in public repositories is frequently curated and updated; therefore the results
of this example might be slightly different when you use up-to-date datasets.
It is also possible that you get warnings about invalid or obsolete IDs due to
an updated Homo sapiens gene annotation file.

m = GO.Terms(end).id;
% gets the last term id
chipgenesCount = zeros(m,1); % a vector of GO term counts for the entire ch
upgenesCount = zeros(m,1); % a vector of GO term counts for up-regulated
for i = 1:length(huGenes)
if isKey(HGmap,huGenes{i})
goid = getrelatives(GO,HGmap(huGenes{i}));
% Update the tally

4-95

Microarray Analysis

chipgenesCount(goid) = chipgenesCount(goid) + 1;
if (any(i == up_geneidx))
upgenesCount(goid) = upgenesCount(goid) +1;
end
end
end

Determine the statistically significant GO terms using the hypergeometric

probability distribution. For each GO term, a p-value is calculated
representing the probability that the number of annotated genes associated
with it could have been found by chance.
gopvalues = hygepdf(upgenesCount,max(chipgenesCount),...
max(upgenesCount),chipgenesCount);
[dummy, idx] = sort(gopvalues);
report = sprintf('GO Term
p-value
counts
definition\n');
for i = 1:10
term = idx(i);
report = sprintf('%s%s\t%-1.5f\t%3d / %3d\t%s...\n',...
report, char(num2goid(term)), gopvalues(term),...
upgenesCount(term), chipgenesCount(term),...
GO(term).Term.definition(2:min(50,end)));
end
disp(report);
GO Term
GO:0003735
GO:0019843
GO:0008135
GO:0000049
GO:0000498
GO:0001069
GO:0033204
GO:0034336
GO:0034583
GO:0034584

4-96

p-value
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000

57
53
55
52
51
51
51
51
51
51

/
/
/
/
/
/
/
/
/
/

counts
definition
162 The action of a molecule that contributes to t
219 Interacting selectively and non-covalently wit
244 Functions during translation by interacting se
220 Interacting selectively and non-covalently wit
213 Interacting selectively and non-covalently wit
213 Interacting selectively and non-covalently wit
213 Interacting selectively and non-covalently wit
213 Interacting selectively and non-covalently wit
213 Interacting selectively and non-covalently wit
213 Interacting selectively and non-covalently wit

Exploring Gene Expression Data

Inspect the significant GO terms and select the terms related to specific
molecule functions to build a sub-ontology that includes the ancestors of the
terms. Visualize this ontology using the biograph function. You can also color
the graphs nodes. In this example, the red nodes are the most significant,
while the blue nodes are the least significant gene ontology terms. Note: The
GO terms returned may differ from those shown due to the frequent update to
the Homo sapiens gene annotation file.
fcnAncestors = GO(getancestors(GO,idx(1:5)))
[cm acc rels] = getmatrix(fcnAncestors);
BG = biograph(cm,get(fcnAncestors.Terms,'name'))
for i=1:numel(acc)
pval = gopvalues(acc(i));
color = [(1-pval).^(1) pval.^(1/8) pval.^(1/8)];
set(BG.Nodes(i),'Color',color);
end
view(BG)
Gene Ontology object with 13 Terms.
Biograph object with 13 nodes and 14 edges.

Finding the Differentially Expressed Genes in Pathways

You can query the pathway information of the differentially expressed genes
from the KEGG pathway database through KEGGs Web Service.
Following are a few pathway maps with the genes in the up-regulated gene
list highlighted:
Cell Cycle
Hedgehog Signaling pathway
mTor Signaling pathway
References

4-97

Microarray Analysis

[1] Pomeroy, S.L., Tamayo, P., Gaasenbeek, M., Sturla, L.M., Angelo, M.,
McLaughlin, M.E., Kim, J.Y., Goumnerova, L.C., Black, P.M., Lau, C., Allen,
J.C., Zagzag, D., Olson, J.M., Curran, T., Wetmore, C., Biegel, J.A., Poggio, T.,
Mukherjee, S., Rifkin, R., Califano, A., Stolovitzky, G., Louis, DN, Mesirov,
J.P., Lander, E.S., and Golub, T.R. (2002). Prediction of central nervous
system embryonal tumour outcome based on gene expression. Nature,
415(6870), 436-442.
[2] Storey, J.D., and Tibshirani, R. (2003). Statistical significance for
genomewide studies. Proc.Nat.Acad.Sci., 100(16), 9440-9445.
[3] Dudoit, S., Shaffer, J.P., and Boldrick, J.C. (2003). Multiple hypothesis
testing in microarray experiment. Statistical Science, 18, 71-103.
[4] Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery
rate: a practical and powerful approach to multiple testing. J. Royal Stat.
Soc., B 57, 289-300.
Suggest an enhancement for this example.

4-98

5
Phylogenetic Analysis
Overview of Phylogenetic Analysis on page 5-2
Building a Phylogenetic Tree on page 5-3
Phylogenetic Tree App Reference on page 5-16

Phylogenetic Analysis

Overview of Phylogenetic Analysis

Phylogenetic analysis is the process you use to determine the evolutionary
relationships between organisms. The results of an analysis can be drawn
in a hierarchical diagram called a cladogram or phylogram (phylogenetic
tree). The branches in a tree are based on the hypothesized evolutionary
relationships (phylogeny) between organisms. Each member in a branch, also
known as a monophyletic group, is assumed to be descended from a common
ancestor. Originally, phylogenetic trees were created using morphology, but
now, determining evolutionary relationships includes matching patterns in
nucleic acid and protein sequences.

5-2

Building a Phylogenetic Tree

Building a Phylogenetic Tree

In this section...
Overview of the Primate Example on page 5-3
Searching NCBI for Phylogenetic Data on page 5-5
Creating a Phylogenetic Tree for Five Species on page 5-6
Creating a Phylogenetic Tree for Twelve Species on page 5-9
Exploring the Phylogenetic Tree on page 5-11

Note For information on creating a phylogenetic tree with multiply aligned

sequences, see the phytree function.

Overview of the Primate Example

In this example, a phylogenetic tree is constructed from mitochondrial DNA
(mtDNA) sequences for the family Hominidae. This family includes gorillas,
chimpanzees, orangutans, and humans.
The following procedures demonstrate the phylogenetic analysis features
in the Bioinformatics Toolbox software. They are not intended to teach the
process of phylogenetic analysis, but to show you how to use MathWorks
products to create a phylogenetic tree from a set of nonaligned nucleotide
sequences.
The origin of modern humans is a heavily debated issue that scientists have
recently tackled by using mitochondrial DNA (mtDNA) sequences. One
hypothesis explains the limited genetic variation of human mtDNA in terms
of a recent common genetic ancestry, implying that all modern population
mtDNA originated from a single woman who lived in Africa less than 200,000
years ago.

Why Use Mitochondrial DNA Sequences for Phylogenetic

Study?
Mitochondrial DNA sequences, like the Y chromosome, do not recombine
and are inherited from the maternal parent. This lack of recombination

5-3

Phylogenetic Analysis

allows sequences to be traced through one genetic line and all polymorphisms
assumed to be caused by mutations.
Mitochondrial DNA in mammals has a faster mutation rate than nuclear
DNA sequences. This faster rate of mutation produces more variance between
sequences and is an advantage when studying closely related species. The
mitochondrial control region (Displacement or D-loop) is one of the fastest
mutating sequence regions in animal DNA.

Neanderthal DNA
The ability to isolate mitochondrial DNA (mtDNA) from palaeontological
samples has allowed genetic comparisons between extinct species and closely
related nonextinct species. The reasons for isolating mtDNA instead of
nuclear DNA in fossil samples have to do with the fact that:
mtDNA, because it is circular, is more stable and degrades slower then
nuclear DNA.
Each cell can contain a thousand copies of mtDNA and only a single copy
of nuclear DNA.
While there is still controversy as to whether Neanderthals are direct
ancestors of humans or evolved independently, the use of ancient genetic
sequences in phylogenetic analysis adds an interesting dimension to the
question of human ancestry.

References
Ovchinnikov I., et al. (2000). Molecular analysis of Neanderthal DNA from
the northern Caucasus. Nature 404(6777), 490493.
Sajantila A., et al. (1995). Genes and languages in Europe: an analysis of
mitochondrial lineages. Genome Research 5 (1), 4252.
Krings M., et al. (1997). Neanderthal DNA sequences and the origin of
modern humans. Cell 90 (1), 1930.
Jensen-Seaman, M., Kidd K. (2001). Mitochondrial DNA variation and
biogeography of eastern gorillas. Molecular Ecology 10(9), 22412247.

5-4

Building a Phylogenetic Tree

Searching NCBI for Phylogenetic Data

The NCBI taxonomy Web site includes phylogenetic and taxonomic
information from many sources. These sources include the published
literature, Web databases, and taxonomy experts. And while the NCBI
taxonomy database is not a phylogenetic or taxonomic authority, it can be
useful as a gateway to the NCBI biological sequence databases.
This procedure uses the family Hominidae (orangutans, chimpanzees,
gorillas, and humans) as a taxonomy example for searching the NCBI Web
site and locating mitochondrial D-loop sequences.
1 Use the MATLAB Help browser to search for data on the Web. In the

MATLAB Command Window, type

web('http://www.ncbi.nlm.nih.gov')

A separate browser window opens with the home page for the NCBI Web
site.
2 Search the NCBI Web site for information. For example, to search for the

human taxonomy, from the Search list, select Taxonomy, and in the for
box, enter hominidae.

The NCBI Web search returns a list of links to relevant pages.

5-5

Phylogenetic Analysis

3 Select the taxonomy link for the family Hominidae. A page with the

taxonomy for the family is shown.

Creating a Phylogenetic Tree for Five Species

Drawing a phylogenetic tree using sequence data is helpful when you are
trying to visualize the evolutionary relationships between species. The
sequences can be multiply aligned or a set of nonaligned sequences, you can
select a method for calculating pairwise distances between sequences, and

5-6

Building a Phylogenetic Tree

you can select a method for calculating the hierarchical clustering distances
used to build a tree.
After locating the GenBank accession codes for the sequences you are
interested in studying, you can create a phylogenetic tree with the data. For
information on locating accession codes, see Searching NCBI for Phylogenetic
Data on page 5-5.
In the following example, you will use the Jukes-Cantor method to calculate
distances between sequences, and the Unweighted Pair Group Method
Average (UPGMA) method for linking the tree nodes.
1 Create a MATLAB structure with information about the sequences. This

step uses the accession codes for the mitochondrial D-loop sequences
isolated from different hominid species.
data = {'German_Neanderthal'
'Russian_Neanderthal'
'European_Human'
'Mountain_Gorilla_Rwanda'
'Chimp_Troglodytes'
};

'AF011222';
'AF254446';
'X90314' ;
'AF089820';
'AF176766';

2 Retrieve sequence data from the GenBank database and copy into the

MATLAB environment.
for ind = 1:5
seqs(ind).Header
= data{ind,1};
seqs(ind).Sequence = getgenbank(data{ind,2},...
'sequenceonly', true);
end
3 Calculate pairwise distances and create a phytree object. For example,

compute the pairwise distances using the Jukes-Cantor distance method

and build a phylogenetic tree using the UPGMA linkage method. Since
the sequences are not prealigned, seqpdist pairwise aligns them before
computing the distances.
distances = seqpdist(seqs,'Method','Jukes-Cantor','Alphabet','DNA');
tree = seqlinkage(distances,'UPGMA',seqs)

5-7

Phylogenetic Analysis

The MATLAB software displays information about the phytree object.

The function seqpdist calculates the pairwise distances between pairs of
sequences while the function seqlinkage uses the distances to build a
hierarchical cluster tree. First, the most similar sequences are grouped
together, and then sequences are added to the tree in descending order
of similarity.
Phylogenetic tree object with 5 leaves (4 branches)
4 Draw a phylogenetic tree.

h = plot(tree,'orient','top');
ylabel('Evolutionary distance')
set(h.terminalNodeLabels,'Rotation',65)

The MATLAB software draws a phylogenetic tree in a Figure window. In

the figure below, the hypothesized evolutionary relationships between the
species is shown by the location of species on the branches. The horizontal
distances do not have any biological significance.

5-8

Building a Phylogenetic Tree

Creating a Phylogenetic Tree for Twelve Species

Plotting a simple phylogenetic tree for five species seems to indicate a number
of monophyletic groups (see Creating a Phylogenetic Tree for Five Species
on page 5-6). After a preliminary analysis with five species, you can add more
species to your phylogenetic tree. Adding more species to the data set will
help you to confirm the observed monophyletic groups are valid.
1 Add more sequences to a MATLAB structure. For example, add mtDNA

D-loop sequences for other hominid species.

data2 = {'Puti_Orangutan'
'Jari_Orangutan'
'Western_Lowland_Gorilla'
'Eastern_Lowland_Gorilla'

'AF451972';
'AF451964';
'AY079510';
'AF050738';

5-9

Phylogenetic Analysis

'Chimp_Schweinfurthii'
'Chimp_Vellerosus'
'Chimp_Verus'

'AF176722';
'AF315498';
'AF176731';

};

2 Get additional sequence data from the GenBank database, and copy the

data into the next indices of a MATLAB structure.

for ind = 1:7
seqs(ind+5).Header
= data2{ind,1};
seqs(ind+5).Sequence = getgenbank(data2{ind,2},...
'sequenceonly', true);
end
3 Calculate pairwise distances and the hierarchical linkage.

distances = seqpdist(seqs,'Method','Jukes-Cantor','Alpha','DNA');
tree = seqlinkage(distances,'UPGMA',seqs);
4 Draw a phylogenetic tree.

h = plot(tree,'orient','top');
ylabel('Evolutionary distance')
set(h.terminalNodeLabels,'Rotation',65)

The MATLAB software draws a phylogenetic tree in a Figure window. You

can see four main clades for humans, gorillas, chimpanzee, and orangutans.

5-10

Building a Phylogenetic Tree

Exploring the Phylogenetic Tree

After you create a phylogenetic tree, you can explore the tree using the
MATLAB command line or the Phylogenetic Tree app. This procedure
uses the tree created in Creating a Phylogenetic Tree for Twelve Species
on page 5-9 as an example.
1 List the members of a tree.

names = get(tree,'LeafNames')
names =
'German_Neanderthal'
'Russian_Neanderthal'

5-11

Phylogenetic Analysis

'European_Human'
'Chimp_Troglodytes'
'Chimp_Schweinfurthii'
'Chimp_Verus'
'Chimp_Vellerosus'
'Puti_Orangutan'
'Jari_Orangutan'
'Mountain_Gorilla_Rwanda'
'Eastern_Lowland_Gorilla'
'Western_Lowland_Gorilla'

From the list, you can determine the indices for its members. For example,
the European Human leaf is the third entry.
2 Find the closest species to a selected species in a tree. For example, find

the species closest to the European human.

[h_all,h_leaves] = select(tree,'reference',3,...
'criteria','distance',...
'threshold',0.6);
h_all is a list of indices for the nodes within a patristic distance of 0.6 to
the European human leaf, while h_leaves is a list of indices for only the

leaf nodes within the same patristic distance.

A patristic distance is the path length between species calculated from
the hierarchical clustering distances. The path distance is not necessarily
the biological distance.
3 List the names of the closest species.

subtree_names = names(h_leaves)

The MATLAB software prints a list of species with a patristic distance to

the European human less than the specified distance. In this case, the
patristic distance threshold is less than 0.6.
subtree_names =
'German_Neanderthal'
'Russian_Neanderthal'

5-12

Building a Phylogenetic Tree

'European_Human'
'Chimp_Schweinfurthii'
'Chimp_Verus'
'Chimp_Troglodytes'
4 Extract a subtree from the whole tree by removing unwanted leaves. For

example, prune the tree to species within 0.6 of the European human
species.
leaves_to_prune = ~h_leaves;
pruned_tree = prune(tree,leaves_to_prune)
h = plot(pruned_tree,'orient','top');
ylabel('Evolutionary distance')
set(h.terminalNodeLabels,'Rotation',65)

The MATLAB software returns information about the new subtree and
plots the pruned phylogenetic tree in a Figure window.
Phylogenetic tree object with 6 leaves (5 branches)

5-13

Phylogenetic Analysis

5 Explore, edit, and format a phylogenetic tree using the Phylogenetic

Tree app.
phytreeviewer(pruned_tree)

The Phylogenetic Tree window opens, showing the tree.

5-14

Building a Phylogenetic Tree

You can interactively change the appearance of the tree using the app.
For information on using this app, see Phylogenetic Tree App Reference
on page 5-16.

5-15

Phylogenetic Analysis

Phylogenetic Tree App Reference

In this section...
Overview of the Phylogenetic Tree App on page 5-16
Opening the Phylogenetic Tree App on page 5-16
File Menu on page 5-18
Tools Menu on page 5-29
Window Menu on page 5-38
Help Menu on page 5-38

Overview of the Phylogenetic Tree App

The Phylogenetic Tree app allows you to view, edit, format, and explore
phylogenetic tree data. With this app you can prune, reorder, rename
branches, and explore distances. You can also open or save Newick or
ClustalW tree formatted files. The following sections give a description of
menu commands and features for creating publishable tree figures.

Opening the Phylogenetic Tree App

This section illustrates how to draw a phylogenetic tree from data in a
phytree object or a previously saved file.
The Phylogenetic Tree app can read data from Newick and ClustalW tree
formatted files.
This procedure uses the phylogenetic tree data stored in the file pf00002.tree
as an example. The data was retrieved from the protein family (PFAM) Web
database and saved to a file using the accession number PF00002 and the
function gethmmtree.
1 Create a phytree object. For example, to create a phytree object from tree

data in the file pf00002.tree, type

tr= phytreeread('pf00002.tree')

The MATLAB software creates a phytree object.

5-16

Phylogenetic Tree App Reference

Phylogenetic tree object with 33 leaves (32 branches)

2 View the phylogenetic tree using the app.

phytreeviewer(tr)

Alternatively, click Phylogenetic Tree on the Apps tab.

5-17

Phylogenetic Analysis

File Menu
The File menu includes the standard commands for opening and closing a
file, and it includes commands to use phytree object data from the MATLAB
Workspace. The File menu commands are shown below.

New Viewer Command

Use the New Viewer command to open tree data from a file into a second
Phylogenetic Tree viewer.
1 From the File menu, select New Viewer.

The Open A Phylogenetic Tree dialog box opens.

5-18

Phylogenetic Tree App Reference

2 Choose the source for a tree.

MATLAB Workspace Select the Import from Workspace options,

and then select a phytree object from the list.
File Select the Open phylogenetic tree file option, click the
Browse button, select a directory, select a file with the extension .tree,
and then click Open. The toolbox uses the file extension .tree for
Newick-formatted files, but you can use any Newick-formatted file with
any extension.

5-19

Phylogenetic Analysis

A second Phylogenetic Tree viewer opens with tree data from the selected
file.

Open Command
Use the Open command to read tree data from a Newick-formatted file and
display that data in the app.
1 From the File menu, click Open.

The Select Phylogenetic Tree File dialog box opens.

2 Select a directory, select a Newick-formatted file, and then click Open. The

app uses the file extension .tree for Newick-formatted files, but you can
use any Newick-formatted file with any extension.
The app replaces the current tree data with data from the selected file.

Import from Workspace Command

Use the Import from Workspace command to read tree data from a phytree
object in the MATLAB Workspace and display the data using the app.
1 From the File menu, select Import from Workspace.

The Get Phytree Object dialog box opens.

5-20

Phylogenetic Tree App Reference

2 From the list, select a phytree object in the MATLAB Workspace.

3 Click the Import button.

The app replaces the current tree data with data from the selected object.

Open Original in New Viewer

There may be times when you make changes that you would like to undo.
The Phylogenetic Tree app does not have an undo command, but you can
get back to the original tree you started viewing with the Open Original
in New Viewer command.
From the File menu, select Open Original in New Viewer.
A new Phylogenetic Tree viewer opens with the original tree.

Save As Command
After you create a phytree object or prune a tree from existing data, you can
save the resulting tree in a Newick-formatted file. The sequence data used to
create the phytree object is not saved with the tree.
1 From the File menu, select Save As.

5-21

Phylogenetic Analysis

The Save Phylogenetic tree as dialog box opens.

2 In the Filename box, enter the name of a file. The toolbox uses the file

extension .tree for Newick-formatted files, but you can use any file
extension.
3 Click Save.

The app saves tree data without the deleted branches, and it saves changes
to branch and leaf names. Formatting changes such as branch rotations,
collapsed branches, and zoom settings are not saved in the file.

Export to New Viewer Command

Because some of the Phylogenetic Tree viewer commands cannot be undone
(for example, the Prune command), you might want to make a copy of your
tree before trying a command. At other times, you might want to compare two
views of the same tree, and copying a tree to a new tool window allows you to
make changes to both tree views independently .
1 Select File > Export to New Viewer, and then select either With

Hidden Nodes or Only Displayed.

A new Phylogenetic Tree viewer opens with a copy of the tree.
2 Use the new figure to continue your analysis.

Export to Workspace Command

The Phylogenetic Tree app can open Newick-formatted files with tree data.
However, it does not create a phytree object in the MATLAB Workspace. If
you want to programmatically explore phylogenetic trees, you need to use the
Export to Workspace command.
1 Select File > Export to Workspace, and then select either With Hidden

Nodes or Only Displayed.

The Export to Workspace dialog box opens.
2 In the Workspace variable name box, enter the name for your

phylogenetic tree data. For example, enter MyTree.

5-22

Phylogenetic Tree App Reference

3 Click OK.

The app creates a phytree object in the MATLAB Workspace.

Print to Figure Command

After you have explored the relationships between branches and leaves in
your tree, you can copy the tree to a MATLAB Figure window. Using a
Figure window lets you use all the features for annotating, changing font
characteristics, and getting your figure ready for publication. Also, from the
Figure window, you can save an image of the tree as it was displayed in the
Phylogenetic Tree app.
1 From the File menu, select Print to Figure, and then select either With

Hidden Nodes or Only Displayed.

The Print Phylogenetic Tree to Figure dialog box opens.

5-23

Phylogenetic Analysis

2 Select one of the Rendering Types.

5-24

Phylogenetic Tree App Reference

Rendering Type

Description

'square' (default)

'angular'

5-25

Phylogenetic Analysis

Rendering Type
'radial'

'equalangle'

5-26

Description

Phylogenetic Tree App Reference

Rendering Type

Description

Tip This rendering type hides the significance of

the root node and emphasizes clusters, thereby
making it useful for visually assessing clusters
and detecting outliers.
'equaldaylight'

Tip This rendering type hides the significance of

the root node and emphasizes clusters, thereby
making it useful for visually assessing clusters
and detecting outliers.

3 Select the Display Labels you want on your figure. You can select from all

to none of the options.

Branch Nodes Display branch node names on the figure.
Leaf Nodes Display leaf node names on the figure.
Terminal Nodes Display terminal node names on the right border.
4 Click the Print button.

5-27

Phylogenetic Analysis

A new Figure window opens with the characteristics you selected.

Print Preview Command

When you print from the Phylogenetic Tree app or a MATLAB Figure
window (with a tree published from the viewer), you can specify setup options
for printing a tree.
1 From the File menu, select Print Preview.

The Print Preview window opens, which you can use to select page
formatting options.

2 Select the page formatting options and values you want, and then click

Print.

5-28

Phylogenetic Tree App Reference

Print Command
Use the Print command to make a copy of your phylogenetic tree after you
use the Print Preview command to select formatting options.
1 From the File menu, select Print.

The Print dialog box opens.

2 From the Name list, select a printer, and then click OK.

Tools Menu
Use the Tools menu to:
Explore branch paths
Rotate branches
Find, rename, hide, and prune branches and leaves.
The Tools menu and toolbar contain most of the commands specific to trees
and phylogenetic analysis. Use these commands and modes to edit and format
your tree interactively. The Tools menu commands are:

5-29

Phylogenetic Analysis

Inspect Mode
Viewing a phylogenetic tree in the Phylogenetic Tree app provides a rough
idea of how closely related two sequences are. However, to see exactly how
closely related two sequences are, measure the distance of the path between
them. Use the Inspect command to display and measure the path between
two sequences.
1 Select Tools > Inspect, or from the toolbar, click the Inspect Tool Mode

icon

The app is set to inspect mode.

5-30

Phylogenetic Tree App Reference

2 Click a branch or leaf node (selected node), and then hover your cursor over

another branch or leaf node (current node).

The app highlights the path between the two nodes and displays the path
length in the pop-up window. The path length is the patristic distance
calculated by the seqpdist function.

Collapse and Expand Branch Mode

Some trees have thousands of leaf and branch nodes. Displaying all the nodes
can create an unreadable tree diagram. By collapsing some branches, you can
better see the relationships between the remaining nodes.
1 Select Tools > Collapse/Expand, or from the toolbar, click the

Collapse/Expand Brand Mode icon

The app is set to collapse/expand mode.

2 Point to a branch.

The paths, branch nodes, and leaf nodes below the selected branch appear
in gray, indicating you selected them to collapse (hide from view).

3 Click the branch node.

The app hides the display of paths, branch nodes, and leaf nodes below the
selected branch. However, it does not remove the data.

5-31

Phylogenetic Analysis

4 To expand a collapsed branch, click it or select Tools > Reset View.

Tip After collapsing nodes, you can redraw the tree by selecting Tools >
Fit to Window.

Rotate Branch Mode

A phylogenetic tree is initially created by pairing the two most similar
sequences and then adding the remaining sequences in a decreasing order of
similarity. You can rotate branches to emphasize the direction of evolution.
1 Select Tools > Rotate Branch, or from the toolbar, click the Rotate

Branch Mode icon

The app is set to rotate branch mode.

2 Point to a branch node.

3 Click the branch node.

5-32

Phylogenetic Tree App Reference

The branch and leaf nodes below the selected branch node rotate 180
degrees around the branch node.
4 To undo the rotation, simply click the branch node again.

Rename Leaf or Branch Mode

The Phylogenetic Tree app takes the node names from a phytree object
and creates numbered branch names starting with Branch 1. You can edit
any of the leaf or branch names.
1 Select Tools > Rename, or from the toolbar, click the Rename

Leaf/Branch Mode icon

The app is set to rename mode.

2 Click a branch or leaf node.

A text box opens with the current name of the node.

3 In the text box, edit or enter a new name.

4 To accept your changes and close the text box, click outside of the text box.

To save your changes, select File > Save As.

Prune (Delete) Leaf or Branch Mode

Your tree can contain leaves that are far outside the phylogeny, or it can have
duplicate leaves that you want to remove.

5-33

Phylogenetic Analysis

1 Select Tools > Prune, or from the toolbar, click the Prune (delete)

Leaf/Branch Mode icon

The app is set to prune mode.

2 Point to a branch or leaf node.

For a leaf node, the branch line connected to the leaf appears in gray. For a
branch node, the branch lines below the node appear in gray.
Note If you delete nodes (branches or leaves), you cannot undo the
changes. The Phylogenetic Tree app does not have an Undo command.
3 Click the branch or leaf node.

The tool removes the branch from the figure and rearranges the other
nodes to balance the tree structure. It does not recalculate the phylogeny.
Tip After pruning nodes, you can redraw the tree by selecting Tools > Fit
to Window.

Zoom In, Zoom Out, and Pan Commands

The Zoom and Pan commands are the standard controls for resizing and
moving the screen in any MATLAB Figure window.
1 Select Tools > Zoom In, or from the toolbar, click the Zoom In icon

5-34

Phylogenetic Tree App Reference

The app activates zoom in mode and changes the cursor to a magnifying
glass.

2 Place the cursor over the section of the tree diagram you want to enlarge

and then click.

The tree diagram doubles its size.

3 From the toolbar click the Pan icon

4 Move the cursor over the tree diagram, left-click, and drag the diagram to

the location you want to view.

Tip After zooming and panning, you can reset the tree to its original view,
by selecting Tools > Reset View.

Select Submenu
Select a single branch or leaf node by clicking it. Select multiple branch or
leaf nodes by Shift-clicking the nodes, or click-dragging to draw a box around
nodes.

5-35

Phylogenetic Analysis

Use the Select submenu to select specific branch and leaf nodes based on
different criteria.
Select By Distance Displays a slider bar at the top of the window,
which you slide to specify a distance threshold. Nodes whose distance from
the selected node are below this threshold appear in red. Nodes whose
distance from the selected node are above this threshold appear in blue.
Select Common Ancestor For all selected nodes, highlights the closest
common ancestor branch node in red.
Select Leaves If one or more nodes are selected, highlights the nodes
that are leaf nodes in red. If no nodes are selected, highlights all leaf
nodes in red
Propagate Selection For all selected nodes, highlights the descendant
nodes in red.
Swap Selection Clears all selected nodes and selects all deselected
nodes.
After selecting nodes using one of the previous commands, hide and show the
nodes using the following commands:
Collapse Selected
Expand Selected
Expand All
Clear all selected nodes by clicking anywhere else in the Phylogenetic Tree
app.

Find Leaf or Branch Command

Phylogenetic trees can have thousands of leaves and branches, and finding a
specific node can be difficult. Use the Find Leaf/Branch command to locate
a node using its name or part of its name.
1 Select Tools > Find Leaf/Branch.

The Find Leaf/Branch dialog box opens.

5-36

Phylogenetic Tree App Reference

2 In the Regular Expression to match box, enter a name or partial name

of a branch or leaf node.

3 Click OK.

The branch or leaf nodes that match the expression appear in red.
After selecting nodes using the Find Leaf/Branch command, you can hide
and show the nodes using the following commands:
Collapse Selected
Expand Selected
Expand All

Collapse Selected, Expand Selected, and Expand All

Commands
When you select nodes, either manually or using the previous commands, you
can then collapse them by selecting Tools > Collapse Selected.
The data for branches and leaves that you hide using the Collapse/Expand
or Collapse Selected command are not removed from the tree. You can
display selected or all hidden data using the Expand Selected or Expand
All command.

Fit to Window Command

After you hide nodes with the collapse commands, or delete nodes with the
Prune command, there can be extra space in the tree diagram. Use the Fit

5-37

Phylogenetic Analysis

to Window command to redraw the tree diagram to fill the entire Figure
window.
Select Tools > Fit to Window.

Reset View Command

Use the Reset View command to remove formatting changes such as
collapsed branches and zooms.
Select Tools > Reset View.

Options Submenu
Use the Options command to select the behavior for the zoom and pan modes.
Unconstrained Zoom Allow zooming in both horizontal and vertical
directions.
Horizontal Zoom Restrict zooming to the horizontal direction.
Vertical Zoom (default) Restrict zooming to the vertical direction.
Unconstrained Pan Allow panning in both horizontal and vertical
directions.
Horizontal Pan Restrict panning to the horizontal direction.
Vertical Pan (default) Restrict panning to the vertical direction.

Window Menu
This section illustrates how to switch to any open window.
The Window menu is standard on MATLAB interfaces and Figure windows.
Use this menu to select any opened window.

Help Menu
This section illustrates how to select quick links to the Bioinformatics
Toolbox documentation for phylogenetic analysis functions, tutorials, and the
Phylogenetic Tree app reference.

5-38

Phylogenetic Tree App Reference

Use the Help menu to select quick links to the Bioinformatics Toolbox
documentation for phylogenetic analysis functions, tutorials, and the
phytreeviewer reference.

5-39