Functional Analysis

of ChIP-Region

Bioinformatics Core Team

(http://mrccsc.github.io/training/)

Overview

• Enrichment for gene sets
• Motif identification

Fisher or Hypergeometric testing

Often used in expression analysis, binomial, hypergeometric and Fisher enrichment tests can be used for ChIP-seq too.

• First annotate peaks to genes
• Compare peaks in genes in gene set to expected number of peaks in genes in gene set.
• The complexity is annotating peaks to the right gene and hence gene set.

Simple approach

• Select a region around TSS and promoter of consistent size.
• Promoter/TSS sizes are forced to be the same size.
• This means no gene has a unfair advantage over other genes to associated to a peak
  • Longer genes
  • Genes in isolation if associated by nearest method.

Simple approach

• A peak has been associated to one or more promoters/TSS it overlaps.
• The gene set test is then performed by binomial/fisher/hypergeometric test as for expression arrays.
• This can be done post Peak to Gene association in programs like David

Less simple approach

• The GREAT piece of software annotates peaks to genes regulatory regions.
• Regulatory regions are created by extending genes until they meet the regulatory region of another gene or a cut-off for distance.
• This means every gene has different chance of having an overlap with a peak.
• The GREAT software then uses a binomial test to compare peaks in gene sets regulatory regions based on the gene sets proportion of the genome.

Other tools

• There are lots of tools for ChIP-seq gene ontology analysis.
• These are just some of the most popular
• Some account for Mappability or use permutation for statistics.

Motif Enrichment

• Motif enrichment analysis can easily be split into two parts
  • Denovo motif identification.
  • Enrichment for known motifs
• Both require a background which is appropriate

Choosing a background

• For expression studies this is quite straight forward
  • All promoters for instance
• For ChIP-seq this isn't so obvious.
• A background can be gained by scanning the whole genome.
• Or from permuting the supplied sequences to maintain sequence structure. 

Choosing a database of Motifs

• Up until recently Transfac pro was the only option.
• But more recently Jaspar database got an update.
• Jaspar is freely available and is maintained as an R package by Ge Tan in Boris's group.

PWMS

• PWMs are point weight matrices.
• They represent the probability of a base occurring at that position for a motif.
• They are the result from Denovo motif searching and input to known motif searching

|   |         1|  2|  3|         4|  5|  6|  7|         8|         9|        10|
|:--|---------:|--:|--:|---------:|--:|--:|--:|---------:|---------:|---------:|
|A  | 0.2213683|  0|  1| 0.0000000|  0|  0|  0| 0.0091846| 0.1589503| 0.0153702|
|C  | 0.6841612|  1|  0| 0.4301781|  0|  0|  0| 0.4305530| 0.2037488| 0.3518276|
|G  | 0.0944705|  0|  0| 0.0000000|  1|  0|  1| 0.3900656| 0.1042174| 0.1501406|
|T  | 0.0000000|  0|  0| 0.5698219|  0|  1|  0| 0.1701968| 0.5330834| 0.4826617|

Denovo Motif Searches

• Denovo searches for denovo motifs and will find motifs most likely to be enriched in your sequence with no prior knowledge.
• Motifs discovered will be compared to a database to see if they are known.
• Often a background of scrambled sequences or Markov models from input are used.
• NestedMica and Meme-ChIP are two popular tools. 

Known Motif Searches

• Search sequences for known motifs.
  • All Motifs from Jaspar
• Will score success sequences against maximum score for PWM and present as proportion.
• When compared between two groups or a suitable background this can be useful.

Time for the last bit of code!