GPAT: Retrieval of genomic annotation from large genomic position datasets
© Krebs et al; licensee BioMed Central Ltd. 2008
Received: 16 September 2008
Accepted: 15 December 2008
Published: 15 December 2008
Recent genome wide transcription factor binding site or chromatin modification mapping analysis techniques, such as chromatin immunoprecipitation (ChIP) linked to DNA microarray analysis (ChIP on chip) or ChIP coupled to high throughput sequencing (ChIP-seq), generate tremendous amounts of genomic location data in the form of one-dimensional series of signals. After pre-analysis of these data (signal pre-clearing, relevant binding site detection), biologists need to search for the biological relevance of the detected genomic positions representing transcription regulation or chromatin modification events.
To address this problem, we have developed a G enomic P osition A nnotation T ool (GPAT) with a simple web interface that allows the rapid and systematic labelling of thousands of genomic positions with several types of annotations. GPAT automatically extracts gene annotation information around the submitted positions from different public databases (Refseq or ENSEMBL). In addition, GPAT provides access to the expression status of the corresponding genes from either existing transcriptomic databases or from user generated expression data sets. Furthermore, GPAT allows the localisation of the genomic coordinates relative to the chromosome bands and the well characterised ENCODE regions. We successfully used GPAT to analyse ChIP on chip data and to identify genes functionally regulated by the TATA binding protein (TBP).
GPAT provides a quick, convenient and flexible way to annotate large sets of genomic positions obtained after pre-analysis of ChIP-chip, ChIP-seq or other high throughput sequencing-based techniques. Through the different annotation data displayed, GPAT facilitates the interpretation of genome wide datasets for molecular biologists.
One of the major issues in genomics is the genome wide mapping of transcription factor binding sites in order to study their function at the scale of the genome. The chromatin immunoprecipitation (ChIP) technique uses antibodies that are specific for a transcription factor or a chromatin modification, to isolate the DNA to which this factor or modified histone is bound in a cell at a given time. The recent appearance of several genome wide analysis techniques, where ChIP is either followed by DNA microarray analysis (ChIP on chip) or coupled to high throughput sequencing (ChIP-seq), made the genome wide mapping of DNA bound factors technically possible. However, these analyses generate tremendous amounts of genomic location data in the form of one-dimensional series of signals.
Recently, efforts have been made to develop academic software to pre-analyse these datasets, (e.g. Mpeak ), in order to locate the signal peaks that correspond to functional elements, such as promoters, enhancers, repressors or insulators. However, once these datasets are cleaned and the significant signals are selected, biologists lack user-friendly tools to search for the biological relevance of the resulting binding site genomic positions.
The web interface of GPAT is programmed using Python (v2.5.1)  running on an Apache  WWW server and forms an interactive layer between the user and the underlying processing applications. In order to increase the speed of data recovery, all the data are stored in a local PostgreSQL (v8.1.11) database . The background processes are programmed in Python and take advantage of the PygreSQL  module to efficiently connect to the PostgreSQL database. The class diagram describing the GPAT object oriented python code is displayed in Additional file 1: GPAT class diagram.
Comparison with existing tools
Existing on-line tools, such as the UCSC genome browser , allow the user to display so-called "custom tracks" and to browse locally defined genomic positions, thus facilitating the manual retrieval of biological information over hundreds of annotation tracks. However, the browser does not allow the batch processing of large numbers of positions. Complementary to this approach, studies based on genome wide analyses require a tool to systematically annotate data in a form suitable for further statistical analysis. To address this problem, several applications have been released recently.
Galaxy  is a framework giving access to popular sources of data, such as the UCSC Table Browser  or Biomart at ENSEMBL , using a variety of integrated tools. Although very powerful for certain applications, it remains too general for our purposes. In particular, the "Fetch the closest feature" module in the "Operate on genomic intervals" section that allows the annotation of genomic positions with gene identifiers has some limitations compared to a more specialized tool, such as GPAT (see below). As an example, the "Fetch the closest feature" module searches for the closest feature without distance limitations, and often matches very distant genes with questionable biological relevance. In addition, the output is limited to a plain text file corresponding to the concatenation of the input files with restricted information content (e.g. the distance to the TSS is not calculated and no hyperlinks to gene annotations are provided). Moreover, the direct cross-linking of the identified gene list with transcriptomic datasets is not directly possible.
CEAS  is a ChIP on chip analysis pipeline available via a web server, which includes a basic genomic position annotation function. One of the major limitations of CEAS is that the annotation search is completely automated and the user cannot adjust the default options. For example, the search window around the submitted position is fixed at 300 bp. This means that the software is not suitable for studies involving factors, which are known to bind enhancer regions located outside of the proximal promoter regions. Furthermore, since the full analysis is done at each dataset submission, the analysis time rapidly increases for large datasets (20 minutes for 1000 positions).
Cisgenome  is a powerful high throughput ChIP analysis package proposing a gene annotation function. However, the user has to perform a full installation process and must store the annotation files on a local computer (several gigabytes) in order to annotate his results. Furthermore, Cisgenome does not provide direct access to other annotations and experimental data (e.g. transcriptomic data), which is a necessary step in the extraction of biological meaning from high throughput data.
The input format is based on the GFF standard file format . Mouse and human data produced by the two latest UCSC genome assembly releases (mm 8, mm 9, hg 17, hg 18) are currently supported.
The "direct search" mode determines whether the submitted positions are detected inside a transcription unit. It was developed to allow searches for a binding site inside the transcription units (e.g. retrovirus insertion events).
In the "window search" mode, a user-defined window is calculated around the submitted positions. Then GPAT searches gene annotations located within this window. It was developed for datasets where no particular binding profile is expected or known (e.g. insulator elements or proteins of unknown function).
The "promoter search" mode uses the same window as the "window search" mode, but tests whether the transcription start site (TSS) of a transcription unit is found within this window. It was developed for datasets, where a binding in the neighbourhood of the TSS is expected (e.g. transcription factors).
Regardless of the search mode chosen by the user, GPAT provides a complete localisation report relative to the matched gene annotation (position of the TSS of the corresponding gene, distance of the detected location to the TSS). Furthermore, when multiple gene matches are found, the user can choose to retrieve either the closest annotation only or all the matched annotations.
Biologists are often interested in the expression status of the genes neighbouring the submitted genomic locations. Therefore, the GPAT software also gives access to the corresponding gene expression levels in several commonly used model cell lines as provided by the GNF Symatlas  and Stembase  (nine human and two mouse cell lines including mouse embryonic stem cells). Furthermore, users can upload their own transcriptomic data during the analysis process, thus allowing the retrieval of gene expression levels for datasets generated under various experimental conditions.
GPAT displays the results as a table containing the annotated positions hyperlinked to the data source (Figure 1C). A direct hyperlink to the UCSC genome browser  is provided to allow the user to browse other genomic features in the case of a successful match. In addition, the full set of results can be downloaded as a tabulated flat file and easily imported in any spreadsheet software for further analysis.
Finally, an additional analysis step allows the user to create a report file, containing summary information, in addition to the matched gene list. This option calculates the global distribution profile of the matched genomic positions as compared to the TSSs of the corresponding genes using parameters set by the user (see example in Figure 1C). Statistical values for the expression data of the corresponding genes are also calculated (Figure 1C).
Use of GPAT – Example:
The Tata Binding Protein (TBP) is a component of a number of complexes, including the TFIID complex involved in the RNA Polymerase II (Pol II) general transcription machinery. Surprisingly, it has been shown that Pol II transcription can occur in the absence of TBP [16, 17]. However, the molecular mechanism leading to TBP-free transcription is still poorly understood. One of the strategies for studying this mechanism is to isolate target genes where Pol II transcription can be detected in the absence of TBP.
We tested GPAT using a promoter DNA tiling array dataset generated by hybridization of DNA (prepared from Hela cells), which was ChIPed using specific antibodies against Pol II, TBP and GST (as a negative control). For each experiment, three slides (Agilent G4483A – 013863, 013864 and 01387) representing approximately one-third of the most characterised promoters of the gene were hybridized (about 5600 genes covered from -8 kb upstream to +2.5 kb downstream with a ~350 bp resolution). After intra-array lowess normalization, peaks corresponding to the factor binding sites were detected using Agilent ChIP analytics (using the neighbourhood model).
Finally, the output of GPAT allowed us to attribute an expression level for each of the differently regulated gene categories showing that the genes for which a co-occupancy of TBP and Pol II was detected have the strongest expression levels (Figure 2C). This correlation gives a high level of confidence in the list of binding sites established by our analysis. Furthermore, by combining these data, we obtained a list of expressed genes, for which Pol II, but not TBP is present (Figure 2A). These genes can be used as a model for further analysis of the mechanism of TBP-free transcription initiation. Finally, the group of genes for which TBP, but not Pol II is bound at the promoter shows low expression levels, implying that there are genes where TBP is bound, but Pol II, is not or is poorly recruited.
In conclusion, these results demonstrate the efficiency of GPAT for the extraction of biological meaning from large genomic position datasets.
The current version of GPAT improves the bioinformatics analysis of ChIP on chip or ChIP-seq data. It considerably speeds up and facilitates the steps between data pre-analysis and the biological interpretation of the data. Notably, the amplitude in the stringency of the annotation search provided by the three different research modes should meet the requirements of most of the biological questions addressed. Furthermore, the full localisation report (including distance to the TSS) will help the biologist to easily understand the binding pattern of his studied factors. Finally, the connection of the binding sites and the gene expression data is an essential resource at different steps of the analysis. Firstly, it provides a supplementary filtering step to distinguish between relevant and background signals. Secondly, it adds another layer of complexity to the data interpretation, providing insight into the regulation processes taking place at particular genomic locations.
Currently, the investigation process of large scale ChIP data is complicated and involves multiple analysis steps. In order to make this technology available for laboratories without access to bioinformatics expertise, considerable efforts are needed to facilitate the data analysis (similar to the efforts dedicated to classical expression arrays). To address this issue, GPAT was designed to improve the steps following data pre-analysis (annotation, cross-linking with other experimental data). However, several improvements remain to be implemented. GPAT currently supports the mouse and human genomes, but adding other model organisms for which genome wide ChIP data are available (e.g. Drosophila) would make the impact of GPAT wider and could be easily implemented in the future.
Furthermore, GPAT provides expression data for several cell lines. One of the recent improvements of the ChIP method allows the use of tissues as a starting material. Since the GNF dataset was also generated for mouse tissues, these datasets could be easily integrated into the GPAT interface.
GPAT provides a quick, convenient and flexible way to annotate large sets of genomic positions obtained after pre-analysis of ChIP on chip, ChIP-seq or other high throughput sequencing based techniques. Thanks to the different annotation and experimental data provided (including the expression status of the identified genes), GPAT facilitates the interpretation of genome wide datasets. We hope that GPAT will be of great help to molecular biologists who wish to analyse large-scale genomic data.
Availability and requirements
Project home page http://bips.u-strasbg.fr/GPAT/Gpat_home.html
The GPAT source code can be freely downloaded from: http://bips.u-strasbg.fr/GPAT/Gpat_source_code.html
For use: Standard WWW browser
GNU General Public License v3 (GPL 3)
- ChIP chip:
Chromatin Immunoprecipitation followed by hybridation on DNA microarray
Chromatin Immunoprecipitation coupled to high throughput Sequencing
Genomic Position Annotation Tool
- RNA Pol II:
RNA polymerase II
TATA Binding Protein
Transcription Start Site
This work was funded by the integrated European Transcriptome, Regulome & Cellular Commitment Consorsium (EUTRACC) project. AK is recipient of a fellowship from the INSERM and the Alsace region.
The authors thank the organisers and teachers of the "Informatics in biology" course of the Pasteur Institute (Paris, France) where GPAT was designed and implemented; M. Vandenbogaert and the IGBMC bioinformatics platform (particularly F. Plewniak) for technical support; C. Lethondal, C. Saveanu for critical reading of the manuscript; J. Thompson for advices and helpful comments on the manuscript.
- Mpeak software[http://www.stat.ucla.edu/~zmdl/mpeak]
- Python (v2.5.1)[http://www.python.org]
- Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, Zahler AM, Haussler D: The human genome browser at UCSC. Genome Res 2002, 12: 996–1006.PubMed CentralView ArticlePubMedGoogle Scholar
- Ji X, Li W, Song J, Wei L, Liu XS: CEAS: cis-regulatory element annotation system. Nucleic Acids Res 2006, 34: W551–554. 10.1093/nar/gkl322PubMed CentralView ArticlePubMedGoogle Scholar
- Ji H, Jiang H, Ma W, Johnson DS, Myers RM, Wong WH: An integrated software system for analyzing ChIP-chip and ChIP-seq data. Nat Biotechnol 2008, 26: 1293–1300. 10.1038/nbt.1505PubMed CentralView ArticlePubMedGoogle Scholar
- GFF Format[http://www.sanger.ac.uk/Software/formats/GFF/GFF_Spec.shtml]
- Su AI, Wiltshire T, Batalov S, Lapp H, Ching KA, Block D, Zhang J, Soden R, Hayakawa M, Kreiman G, Cooke MP, Walker JR, Hogenesch JB: A gene atlas of the mouse and human protein-encoding transcriptomes. Proc Natl Acad Sci USA 2004, 101: 6062–6067. 10.1073/pnas.0400782101PubMed CentralView ArticlePubMedGoogle Scholar
- Perez-Iratxeta C, Palidwor G, Porter CJ, Sanche NA, Huska MR, Suomela BP, Muro EM, Krzyzanowski PM, Hughes E, Campbell PA, Rudnicki MA, Andrade MA: Study of stem cell function using microarray experiments. FEBS Lett 2005, 579: 1795–1801. 10.1016/j.febslet.2005.02.020View ArticlePubMedGoogle Scholar
- Furey TS, Haussler D: Integration of the cytogenetic map with the draft human genome sequence. Hum Mol Genet 2003, 12: 1037–1044. 10.1093/hmg/ddg113View ArticlePubMedGoogle Scholar
- ENCODE Project[http://www.genome.gov/10005107]
- Martianov I, Viville S, Davidson I: RNA polymerase II transcription in murine cells lacking the TATA binding protein. Science 2002, 298: 1036–1039. 10.1126/science.1076327View ArticlePubMedGoogle Scholar
- Muller F, Tora L: The multicoloured world of promoter recognition complexes. Embo J 2004, 23: 2–8. 10.1038/sj.emboj.7600027PubMed CentralView ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.