• Supplementary information to our paper: 'Utilizing promoter pair orientations for HMM-based analysis of ChIP-chip data' by Michael Seifert, Jens Keilwagen, Marc Strickert, and Ivo Grosse at the GCB2008 in Dresden.

• GCB2008 talk
• GCB2008 paper (In Proc. of German Conference on Bioinformatics (GCB2008), LNI, Dresden, Germany, September 2008)

• Motivation

  • Seed development of Arabidopsis thaliana as model system for the comparison of different methods of ChIP-chip data analysis

• Downloads

  • ChIPchipAnalyzer: A JAR-file for analyzing ChIP-chip data

    • Log-Fold-Change analysis (LFC)

    • Hidden Markov Model (HMM)

    • Hidden Markov Model with scaled transition matrices (SHMM)

  • ABI3 ChIP-chip data set

• Case Study

  • Shows how to use the ChIPchipAnalyzer in combination with the ABI3 data set.

• General Usage

  • Shows how a data set must be structured to apply the ChIPchipAnalyzer.

• ChIP-chip is still far from being routinely used for Arabidopsis thaliana

• ChIP-chip based on promoter arrays was established for the transcription factor ABI3 in the trilateral project ArabidoSeed

• ABI3 is a fundamental regulator of seed development

• Knowledge of ABI3 target genes is important for a better understanding of seed development

• One of the bioinformatics tasks is to provide tools for the detection of these target genes

  1. Log-Fold Change analysis (LFC)

  2. Two-State Hidden Markov Model (HMM) modeling chromosomal locations of promoters

  3. Two-State Hidden Markov Model with scaled transition matrices (SHMM) modeling chromosomal locations of promoters and promoter pair orientations

• Considering ChIP-chip data of ABI3 the SHMM can be seen as the more general model that should be preferred for the detection of ABI3 target genes (validation by independent Genevestigator expression data and transient assays)

• We conjecture that SHMMs might possibly be useful for the analysis of other promoter array ChIP-chip data

• ChIPchipAnalyzer (231 KB)

  • Implements LFC, HMMs, and SHMMs to analyze ChIP-chip data.

  • Requires Java 1.5 or higher.

  • Usage information

    java -jar ChIPchipAnalyzer.jar

• ABI3 data set (Currently not possible because the work on a biological paper is sill in progress.) (1,2 MB)

  • ABI3 experiments that we used in our case study.

• Download the ChIPchipAnalyzer and the ABI3 data set to a folder CaseStudy on your system.

• Unpack the ABI3 data set

• Ensure to have enough heap space for Java -Xms256m -Xmx256m

Log-Fold-Change Analysis

• Go to the directory CaseStudy

• Start the ChIPchipAnalyzer

java -jar ChIPchipAnalyzer.jar -LFC -output true -dataSets 061220ABI3.txt 070222ABI3.txt 070223ABI3.txt 070420ABI3.txt 070427ABI3.txt

• You obtain ranking lists in the directory CaseStudy based on decreasing log-ratios for each experiment.

• 061220ABI3.txt_ScoringLFC.txt

• 070222ABI3.txt_ScoringLFC.txt

• 070223ABI3.txt_ScoringLFC.txt

• 070420ABI3.txt_ScoringLFC.txt

• 070427ABI3.txt_ScoringLFC.txt

Standard HMMs

• Go to the directory CaseStudy

• Start the ChIPchipAnalyzer

java -jar ChIPchipAnalyzer.jar -startDistribution 0.9 0.1 -stateDurationScalingFactor 0.05 -means 0.0 2.0 -stds 1.0 0.5 -ess 4 -aprioriMeans 0.0 2.0 -scaleOfAprioriMeans 1000 1000 -shapeOfStandardDeviations 1 100 -scaleOfStandardDeviations 1E-4 1E-4 -output true -dataSets 061220ABI3.txt 070222ABI3.txt 070223ABI3.txt 070420ABI3.txt 070427ABI3.txt -outputFile HMM.txt

• You obtain the trained standard HMM in the directory CaseStudy.

  • HMM.txt

• You obtain ranking lists in the directory CaseStudy based on decreasing state posteriors of state '+' (target state) for each experiment.

• 061220ABI3.txt_ScoringHMM.txt

• 070222ABI3.txt_ScoringHMM.txt

• 070223ABI3.txt_ScoringHMM.txt

• 070420ABI3.txt_ScoringHMM.txt

• 070427ABI3.txt_ScoringHMM.txt

SHMMs

• Go to directory CaseStudy

• Start the ChIPchipAnalyzer

java -jar ChIPchipAnalyzer.jar -startDistribution 0.9 0.1 -stateDurationScalingFactor 0.05 -scalingFactor 4.0 -distanceThreshold 9000 -means 0.0 2.0 -stds 1.0 0.5 -ess 4 -aprioriMeans 0.0 2.0 -scaleOfAprioriMeans 1000 1000 -shapeOfStandardDeviations 1 100 -scaleOfStandardDeviations 1E-4 1E-4 -output true -dataSets 061220ABI3.txt 070222ABI3.txt 070223ABI3.txt 070420ABI3.txt 070427ABI3.txt -outputFile SHMM.txt

• You obtain the trained SHMM.

  • SHMM.txt

• You obtain ranking lists in the directory CaseStudy based on decreasing state posteriors of state '+' (target state) for each experiment.

• 061220ABI3.txt_ScoringSHMM4.0_9000.txt

• 070222ABI3.txt_ScoringSHMM4.0_9000.txt

• 070223ABI3.txt_ScoringSHMM4.0_9000.txt

• 070420ABI3.txt_ScoringSHMM4.0_9000.txt

• 070427ABI3.txt_ScoringSHMM4.0_9000.txt

• The ChIPchipAnalyzer can be applied to data sets structured in the following manner:

1. Headline with columns ID, Chr, Start, End, Strand, NextIsDirectNeighbor, and Log-Ratio

   • ID: Gene Identifier (Gene is represented by its promoter on the array)

   • Chr: Chromosome where the gene is located

   • Start: Start position of the gene

   • End: End position of the gene

   • Strand: Strand of the gene (1: Forward Strand; -1: Reverse strand)

   • NextIsDirectNeighbor: Is the next gene directly adjacent on the chromosome (TRUE or FALSE)?

   • Log-Ratio: Log-Ratio of TF vs. Control

2. Two ID's are in head-head orientation when the NextIsDirectNeighbor attribute of the first ID is TRUE and the strand attributes are -1 (first ID) followed by 1 (second ID).

1. The column delimiter must be the tabulator.

3. No missing values.

4. All rows in the data set file must be sorted first by increasing values for Chr and second by increasing values for Start.