Chip-Seq (IP-vs-Control and Mock-vs-Control) or IP-vs-Mock for peak calling

Question

[microfuge]

Dear All,

I would be really grateful for your advice in this regard. In Chip-Seq where there are all 3 IP, Input and Mock how to get the best peak calls.

• To run IP vs Mock and get peak calls
• To have two runs IP vs Input and Mock vs Input and cancel out the peaks in mock with either bedtools or diffBind.

My hunch and preference would be the second one. As mock is not a random sample from genome and I presume peak callers would need a random sample in control. The fingerprint plots of Mock clearly indicate a diffuse but substantial enrichment similar but less than IP. Also there could be loss of power if (again I presume) if both case and control have enrichment.

I am using the Encode Chip-Seq pipeline and I don't think mock and input can be incorporated together with IP in one go.

So Many Thanks!

[ATpoint]

I personally find mock IPs meaningless unless the antibody is super unspecific since you need a lot of PCR amplification to even get a library that can be sequenced. Hence you have quite a biased library. Therefore my vote is IP vs input. Without IP replicates the result is exploratory anyway since you cannot assess how reproducible peaks are.

[microfuge]

Thanks! We have 3 replicates of each and I was surprised to see that mock has quite some enrichment although not focal so is for IP and library complexity is not very different between Mock and IP nor is the duplicate fraction. There is substantial peak overlap between (IPvsControl and MockvsControl). I will go ahead with the second option then. Special Thanks for the "unless the antibody is super unspecific" tip.

[ATpoint]

The basic idea of using the mock to detect false peaks is not wrong. Still, you would need a dedicated differential analysis because even if you can call some peaks with mock they can still be much more significant (and therefore true) in the IP. I would even expect that because some unspecific binding of the mock antibody (probably an unspecific IgG or something similar) is expected especially to open chromatin regions where all kinds of regulatory proteins accumulate. I assume though that the overall distribution between IP and mock is quite different so normalization with the standard pipelines like edgeR/csaw is probably not straightforward since you do expect a global change in peak distribution. Meaning a lot of peaks in IP and almost none in mock, and this typically does not go together with the assumptions of most differential analysis tools, which is either that most regions do not change or changes are at least somewhat symmetric. Others might have a better suggestion but I would probably only use the input. As I said the mock probably required a lot of PCR compared to the IP (right?) so it probably is quite biased and comparability to the IP is probably limited.

Edit: It has been a while since I read the manual of the csaw package the last time, check it out maybe it has some good suggestions, and it is tailored for ChIP-seq.