I have a ChIP-Seq dataset with two biological replicates of a treatment (TRT), and two of a control (VEH). All four samples were IPed with a TF antibody, and with IgG as a control. I have mapped all files (control, treatment x 2 reps x IgG and TF) to the hg38 genome, and called narrowpeaks using MACS (v3). I'm looking at the binding sites of the TF under the two conditions, and seeing if the treatment affects how this TF binds.
I have tried using DiffBind in various modes and there is no differential binding between the two conditions (TRT v VEH) and the signal fold changes are not significant.
However, looking at the peaks separately, there is no overlap between the two TRT replicates in terms of peaks, but the two VEH replicates are quite close.
What strategies are there for performing some sort of differential analysis with this dataset?
I did think that you could pool the two biological replicates (but there's no overlap), and do TRT (combined) v VEH (combined), or do a TRT (combined) v VEH (separate), but with effectively one sample, there would be no estimate of significance besides looking at raw signal fold changes.
Ideally, I'd get the experiment done again, but the results are from a few years ago, so it would take a bit of time to spin up and redo.
DiffBind throws an error with a single replicate, which is why I've not run it that way.
But MANorm's permutation test might just be what I'm looking for in terms of a test to run with pooled replicates.