Is it acceptable to use different FDR cutoffs for peak calling in different ChIP-seq experiments (with same antibody)?

Question

[science_lizard]

This is more of a statistics question, but I'm wondering if you think that it's acceptable to use different FDR thresholds for peak calling in one experiment vs. another with the same antibody.

I have some ChIP files (this is actually not ChIP, but it's a protocol very similar except that I use restriction enzymes to digest the DNA instead of sonication or regularly-spaced MNase digestion etc) from human brain, and then I have some also in iPSC models. After I call peaks with HOMER with a stringent FDR threshold (0.0001) on the iPSC models and check them out in IGV, I find that the peaks called are decent based on examination of the BAMs with the peak calls. However, for some reason, even with the same protocol, the peaks called on the in vivo data at that FDR are not so great (seems to be too stringent), and if I look at the restriction enzyme sites I can also see that the peaks called are not even reaching where they should based on where the DNA was fragmented. This starts to go away when I loosen the FDR threshold and I start seeing the peaks called make much more sense based on the RE sites and the BAM file coverage, but then at 'looser' FDR cutoffs for the in vitro data starts to call peaks on a lot of background. So in short, is this acceptable, to use two different FDR thresholds for the same technique?

Thanks!

Answer 1

[benformatics]

FDR thresholds are somewhat arbitrary. Just because people often use 0.05 or 0.01 doesn't mean that they are ideal. Your in vivo data is probably lower quality than the cell line so I think that lowering your threshold is perfectly fine. Unless you have reason to believe that the latter peaks are all just background noise.

[science_lizard]

That's kind of what I was thinking but I wanted to make sure this wasn't breaking some cardinal rule of peak calling. Thank you!!