CASSys: an integrated software-system for the interactive analysis of ChIP-seq data.

Abstract

The mapping of DNA-protein interactions is crucial for a full understanding of transcriptional regulation. Chromatin-immunoprecipitation followed by massively parallel sequencing (ChIP-seq) has become the standard technique for analyzing these interactions on a genome-wide scale. We have developed a software system called CASSys (ChIP-seq data Analysis Software System) spanning all steps of ChIP-seq data analysis. It supersedes the laborious application of several single command line tools. CASSys provides functionality ranging from quality assessment and -control of short reads, over the mapping of reads against a reference genome (readmapping) and the detection of enriched regions (peakdetection) to various follow-up analyses. The latter are accessible via a state-of-the-art web interface and can be performed interactively by the user. The follow-up analyses allow for flexible user defined association of putative interaction sites with genes, visualization of their genomic context with an integrated genome browser, the detection of putative binding motifs, the identification of over-represented Gene Ontology-terms, pathway analysis and the visualization of interaction networks. The system is client-server based, accessible via a web browser and does not require any software installation on the client side. To demonstrate CASSys's functionality we used the system for the complete data analysis of a publicly available Chip-seq study that investigated the role of the transcription factor estrogen receptor-α in breast cancer cells.

Full text

Page 1

CASSys: an integrated software-system for the interactive
analysis of ChIP-seq data
Malik Alawi1,2, Stefan Kurtz1and Michael Beckstette1,*
1Center for Bioinformatics, University of Hamburg, Bundesstraße 43, D-20146 Hamburg,
Germany, http://www.zbh.uni-hamburg.de
2Hamburg Center for Experimental Therapy Research (HEXT), University Medical Center
Hamburg-Eppendorf, Martinistraße 52, D-20246 Hamburg, Germany, http://www.uke.de
Summary
The mapping of DNA-protein interactions is crucial for a full understanding of transcrip-
tionalregulation. Chromatin-immunoprecipitationfollowedbymassivelyparallelsequenc-
ing (ChIP-seq) has become the standard technique for analyzing these interactions on a
genome-wide scale. We have developed a software system called CASSys (ChIP-seq data
Analysis Software System) spanning all steps of ChIP-seq data analysis. It supersedes the
laborious application of several single command line tools. CASSys provides functionality
ranging from quality assessment and -control of short reads, over the mapping of reads
against a reference genome (readmapping) and the detection of enriched regions (peakde-
tection) to various follow-up analyses. The latter are accessible via a state-of-the-art web
interface and can be performed interactively by the user. The follow-up analyses allow for
flexible user defined association of putative interaction sites with genes, visualization of
their genomic context with an integrated genome browser, the detection of putative bind-
ing motifs, the identification of over-represented Gene Ontology-terms, pathway analysis
and the visualization of interaction networks. The system is client-server based, accessi-
ble via a web browser and does not require any software installation on the client side.
To demonstrate CASSys’s functionality we used the system for the complete data analysis
of a publicly available Chip-seq study that investigated the role of the transcription factor
estrogen receptor-α in breast cancer cells.
1Introduction
ChIP-seq, chromatin immunoprecipitation followed by massively parallel sequencing, is a
method for profiling DNA-protein interactions on a genome-wide scale. Offering a higher
resolution and benefiting from the decreasing costs of second-generation sequencing, ChIP-
seq is replacing its array-based predecessor ChIP-chip as the method of choice for identifying
transcription factor binding sites and histone modifications. In a typical ChIP-seq experiment,
immunoprecipitated DNA fragments, about 200 bp long, are sequenced from both ends with
modern short read sequencing technology. As a result, millions of short sequence reads with a
length of 25 - 100 bp are obtained. These short reads are the starting point for the subsequent
computational data analysis. Figure 1 shows, that the computational analysis of ChIP-seq ex-
periments consists of four sequential steps. The first step is usually the quality assessment and
*To whom correspondence should be addressed. Email: beckstette@zbh.uni-hamburg.de
Copyright 2011 The Author(s). Published by Journal of Integrative Bioinformatics.
This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License (http://creativecommons.org/licenses/by-nc-nd/3.0/).
Journal of Integrative Bioinformatics, 8(2):155, 2011http://journal.imbio.de
doi:10.2390/biecoll-jib-2011-1551

Page 2

control (QA/QC) of raw reads. Reads satisfying the applied quality criteria are then mapped
against a reference genome (readmapping) and from these mappings the locations of read en-
riched regions are determined to identify sites where the immunoprecipitated protein and DNA
interact (peakdetection). These are called interaction sites in the following. To understand
mechanisms of transcriptional regulation it is not sufficient to know the mere locations of in-
teraction sites. Therefore, subsequently to peakdetection, further steps, like the analysis of
protein-protein interactions, are required. These follow-up analyses are diverse and often re-
quire the integration of genomic annotation data and other information.
ChIP
Sequencing
Quality
assessment
and -control
Readmapping
Peakdetection
Follow-up
analyses
Computational analysis
Laboratory analysis
Figure 1: The typical workflow of a ChIP-seq experiment can be divided into a laboratory and
a computational part. The laboratory part covers chromatin-immunoprecipitation (ChIP) and
massively parallel sequencing. The computational data analysis part consist of four sequentially
performed steps. Namely, (1) the quality control and assessment of short reads, (2) their mapping
against a reference genome and (3) the derivation of putative interaction sites. In a final step (4)
various follow-up analyses to elucidate the biological implications of identified interaction sites can
follow.
For readmapping and peakdetection several programs have been developed, for a recent review
see [33, 24]. While the readmapping tools are used in several other applications of short read
sequencing, the peakdetection tools are specific to ChIP-seq data analysis. Most of these pro-
grams are command line tools. Stand-alone software is also available for somefollow-up analy-
ses. Especially for the detection of sequence motifs several established programs [5, 21, 23, 25]
exist. Standard genome browsers, like the UCSC Genome Browser [17], can be used to visual-
ize binding sites within their genomic context. But overall, computational analysis of ChIP-seq
data remains a patchwork of manual application of different tools.
Some software packages simplify the computational analyses of ChIP-seq data by providing
graphical interfaces for peakdetection or integrating follow-up analyses. W-ChIPeaks [19] and
ChIP-Seq web server [1] are examples for peak-detection tools with a web-based user interface.
Sole-Search [8] additionally provides basic statistics on interaction sites and identifies the genes
closest to each site. The most advanced system in this field is CisGenome [15], which also
integrates a genome browser and provides motif detection capability. The CisGenome GUI
can, however, only be used on MS Windows platforms. MICSA [9] employs FindPeaks [11]
for peakdetection and aims at enhancing its results by removing putative interaction sites not
bearing certain sequence motifs. Although the tools mentioned above can simplify and improve
certain aspects of ChIP-seq analysis, none of them covers all steps of the analysis, see Table 1.
With CASSys (ChIP-seq Analysis Software System) we present a software system integrating
all steps of computational ChIP-seq data analysis.
Copyright 2011 The Author(s). Published by Journal of Integrative Bioinformatics.
This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License (http://creativecommons.org/licenses/by-nc-nd/3.0/).
Journal of Integrative Bioinformatics, 8(2):155, 2011http://journal.imbio.de
doi:10.2390/biecoll-jib-2011-1552

Page 3

W-ChIPeaksChIP-Seq Web Server Sole-SearchMICSACisGenomeCASSys
√
√1,2
√3,4
√5,6
√7
√
√
√
√
√
√
Quality assessment and control
Readmapping
Peakdetection
Motif detection
Motif comparison
Genome browser
Over-representation analysis
Pathway analysis
Interaction analysis
Web interface
Incorporation of genomic annotations
√√√√
√8
√4
√5
√
√
√
√√
√√
1Bowtie [20],2BWA [22],3MACS [34],4FindPeaks [11],5MEME [5],6Weeder [25],7Tomtom [13],8flexmodule
Table 1: Most ChIP-seq analysis pipelines focus on providing a graphical user-interface for
peakdetectionandhardlyofferfurther analysis capabilities followingthis step. ExceptforCASSys,
nosystemsupportsreadmappingaswellaspeakdetection, whichareabsolutelynecessaryforanal-
ysis of ChIP-seq data. In addition, CASSys offers an integrated QA/QC processing step and the
widest range of possible follow up analyses. Hence, it supports the whole computational data anal-
ysis ranging from the processing of raw reads to follow up analyses like motif detection, pathway
analysis and interaction network visualization.
2 Methods
CASSys divides the computational analyses into two parts. The first part spans the steps from
QA/QC to peakdetection. Since these steps are sequentially executed and—once started—do
not require user-interaction, we refer to them as asynchronous analyses in the following. Re-
sults of the asynchronous part are stored in a local database from where they can be efficiently
retrieved for the interactive analyses subsequent of the peakdetection step. For persistent stor-
age CASSys employs the open source object-relational database system PostgreSQL. For an
overview of CASSys’s system architecture see Figure 2.
2.1Asynchronous analyses
To determine the location of interaction sites CASSys employs the concept of workflows. Here a
workflow describes the tools and parameters used for QA/QC, readmapping and peakdetection.
These sets of parameters are stored in the database. The workflow concept has the following
benefits:
• A workflow completely documents the different steps of the asynchronous analysis al-
lowing to fully reproduce them.
• Parts of a workflow can be (re)used in different combinations. If, for example in two
workflows, only the parametersets for peakdetection differ, it is not required to repeat the
other two steps which came before peakdetection.
The QA/QC step starts with the parsing of short reads in the widespread FastQ-format. A filter
removes prefixes, suffixes or whole reads according to user-specified quality and length param-
eters. The mean quality and base-frequency of each position common to all reads is assessed
before and after filtering. The results of this step are summarized, stored in the database and
can be displayed and visualized in the systems web interface.
Copyright 2011 The Author(s). Published by Journal of Integrative Bioinformatics.
This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License (http://creativecommons.org/licenses/by-nc-nd/3.0/).
Journal of Integrative Bioinformatics, 8(2):155, 2011http://journal.imbio.de
doi:10.2390/biecoll-jib-2011-1553

Page 4

Local
database
Web-based analyzes
VisualizationMotif detection
Gene set
analyses
Association of
interaction sites
with genes
Asynchronous workflow
Quality
assessment
and -control
Readmapping
Peakdetection
Pathway
information
(KEGG)
Protein-protein
interaction
(IntAct)
General
annotation
in GFF3
Data from external sources
Experiment
descriptions
(MiniML)
Sequence
motifs
(JASPAR)
Functional
classification
(GO)
Figure 2: An overview of CASSys’s system architecture. The asynchronous part (red) spans the
computational analysis steps up to peakdetection and the parsing of data from external sources.
Refined data resulting from these processes is persistently stored in a local database, and can be
accessed in the web-based analyses of the interactive part (green).
CASSys employs a selection of widely used software tools for readmapping and peakdetec-
tion. Moreover it offers the infrastructure to quickly integrate new tools. This allows users
to choose from different methods for these central processing steps depending on data quality
and experiment design. In particular, different peak-calling algorithms show severe variations
in the obtained results on different datasets [31] which may lead to dramatic changes in the
drawn biological conclusions [18]. Currently, for readmapping CASSys integrates Bowtie [20]
and BWA [22], while for peakdetection the system employs MACS [34] and FindPeaks [11].
Besides determining interaction sites, the asynchronous part covers the parsing of annotation
data. This process is required only once for each studied species. The annotations provide a
biological context for interpreting interaction sites. Genomic annotations can be provided in
GFF3-format [2]. In addition, CASSys parses other annotation from IntAct [3] and KEGG [16]
in the respective formats. Experiment descriptions are imported in MiniML (MIAMI Notation
In Markup Language) format. This allows to easily import complete experiments with their as-
sociated data files from NCBI’s Gene Expression Omnibus (GEO) [6] which, as of April 2011
contains MiniML-files for over 570 studies.
In ChIP-seq data analysis the identification of genes that are co-regulated by the immunopre-
cipitated protein is of central interest. This requires to associate interaction sites identified in
the peakdetection step with annotated genes. To address this, CASSys calculates, based on the
parsed annotation data and the detected putative interaction sites, the distance between the cen-
ter of each site and the closest transcription start of a gene occurring upstream or downstream
on the forward or the reverse strand. In this way, every site is initially associated with up to
Copyright 2011 The Author(s). Published by Journal of Integrative Bioinformatics.
This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License (http://creativecommons.org/licenses/by-nc-nd/3.0/).
Journal of Integrative Bioinformatics, 8(2):155, 2011http://journal.imbio.de
doi:10.2390/biecoll-jib-2011-1554

Page 5

four genes. CASSys also provides the option to only include functional genes when associating
interaction sites and genes. A gene is considered to be functional if and only if it is listed in
the KEGG database. This allows to ignore pseudogenes and other non-coding elements when
determining the set of site/gene associations. As a consequence, up to eight genes can be as-
sociated with each interaction site. This number results from the combination of the features
upstream/downstream, forward/reverse and functional/not necessarily functional. For each in-
teraction site CASSys considers these (up to) eight genes to be the most likely regulated genes
and stores them in the local database. Observe that this precalculated set of interaction site/gene
associations only constitutes the set of all possible associations from which in the interactive
analyses subsets can be generated based on user defined criteria. This precalculation optimizes
access times and avoids redundant calculation during the interactive analyses.
2.2 Interactive analyses
The analyses subsequent to the peakdetection step are interactive and accessible via CASSys’
web interface. This is designed to be used by experimentalists. The parameters of all software
tools used in the interactive analyses can be defined in the web interface. As the different
analysis steps are very fast and all data is retrieved from the local database, the effects of
changing parameters are instantly visible.
CASSys supports the following three types of web-based analyses which are also depicted in
Figure 3:
• The derivation and analysis of sets of candidate genes (gene set analyses).
• The detection and interpretation of sequence motifs in interaction sites (motif detection).
• The visualization and comparison of interaction sites in their genomic context (genome
browsing).
The analyses rely on information which was stored in the database during the asynchronous
part of CASSys. As CASSys uses a local database to connect the asynchronous and interactive
analyses, it does not require access to third party databases at runtime.
2.2.1Motif detection
If the detected sites interact with the same protein, they likely share common binding motifs.
Therefore, CASSys allows interaction sites to be screened for common motifs using the estab-
lished motif detection programs Weeder [25] and MEME [5]. Weeder provides an enumerative
approach to motif discovery, while MEME is based on a probabilistic model. Results are post-
processed and a report with summary statistics including sequence logos of each detected motif
is generated. The motifs can automatically be compared to known motifs from JASPAR [26]
using the Tomtom [13] program. Pairwise alignments of high-ranking database motifs and de-
tected motifs are reported along with corresponding E-values.
Copyright 2011 The Author(s). Published by Journal of Integrative Bioinformatics.
This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License (http://creativecommons.org/licenses/by-nc-nd/3.0/).
Journal of Integrative Bioinformatics, 8(2):155, 2011 http://journal.imbio.de
doi:10.2390/biecoll-jib-2011-1555