ChIP-Seq Integration & Interpretation

[aquleaf]

Hi. I have three ChIP-Seq data for the same sample. Peak calling was done separately. How to combine the peaks of these three experiments together?

One more question, How to associate genes with these binding sites?

Thanks in advance!

[mudshark]

these are technical replicates, right? So you could simply do the peak calling on the combined data. If you want to do the peak calling first the easiest - maybe not the best - way is to define an overlapping set of peaks.

About the association, there are some tools around that can help you (i guess Galaxy). On the other hand that is not a too complicated job, isn't it? Calculating the distances to some gene features and stuff.

[ETHANol]

I think this is what you are looking for:

http://fasta.bioch.virginia.edu/cshl/stubbs/galaxy_bed2genes.html

There are some other ideas on my blog at:

http://ethanomics.wordpress.com/

On the first question, I agree combining the data is a good way to go. You can also get the common peaks between the samples by using the BedTools intersectBed program. It is also available on Galaxy if you are not comfortable with the command line.

[rory]

Are the replicates purely technical (mutliple lanes of the same library) or was the experiment repeated three times (three different libraries)? This makes a big difference. If it is just three lanes of the same library, then I agree the data should be merged and peaks called once. However without replicating the experiment, you will never be able to get reliable results -- you have no clue as to the variation inherent in the experiment, and whether your library represents an outlier.

If the experiment was repeated three times, then you should not combine the reads, but rather work with the three peaksets. Start by checking their overlaps (3-way venn diagram). To minimize false positives, you could take only the peaks that all three replicates agree on, or be less stringent and take those that overlap in at least two of the three libraries. Another approach is to take all the peaks, then go back to the aligned reads and re-score each peak for each replicate (whether or not it was identified as a peak in that replicate).

See also this paper for ideas on how to combine peaksets: http://projecteuclid.org/DPubS?servi...oas/1318514284

If you are using R and Bioconductor, there are a number of packages that can help with this type of analysis. For example DiffBind helps with overlap and re-scoring, and ChIPpeakAnno will map peaks to genes.