Does differential binding analysis across different ChIP-Seq samples makes sense ?

Question

[ATRX]

Hello All,

I am interested to know if the differential binding analysis across different biological conditions makes sense or not. For example, if there are 2 different samples say A and B, where A is IPed using antibody P1 and B is IPed using antibody P2. and have their respective inputs. One does the peak calling using the inputs and get the set of peak for both the samples. Now, you find that a peak at certain genomic loci is present in both A and B; however, looking at the IGV plot you find that the peak in A has more reads than peak in B (considering that all the samples are normalized). When you do the differential analysis you find that the enrichment of the peak is significant in A more than B. Based on the above scenario, I have few questions:

1) I would like to know if such differential analysis makes sense or not ? 2) How can a peak be quantified same as the peak in RNA-Seq ? In RNA-Seq, you quantify the peaks because biologically, one knows there be can multiple copies of same mRNA

Ideally, I think a peak that is present should be considered as 1 and 0 if not. Please correct me if I wrong.

Answer 1

[Devon Ryan]

While possible, such an analysis shouldn't be performed naively. For example, if IPs with P1 are more efficient than those with P2 then one would expect all P1 peaks to be higher...but in a biologically meaningless way. Assuming you can get around that issue, just use diffbind for the differential binding analysis, it shows in its tutorial how to quantify peaks.

BTW, you should not be comparing presence/absence of peaks using a binary label, that's like saying a p-value of 0.49999999 and 0.5 are significantly different.

[ATRX]

Thanks that was helpful!

[goodez]

This post is rather old, but I stumbled across it and thought it was an interesting question. You can actually deal with the issue of different IP efficiencies if you use spike-in chromatin from another organism that's somewhat closely related. This paper describes the method and we're currently using it in my lab.