Suppose that I have three conditions to compare in DiffBind, with two replicates per condition. Suppose that the antibody is the same, but one is WT and the other two - mutations. All pairwise comparisons are needed. Say that in each of the conditions there is a different average number of peaks (say 400 peaks for WT samples, 400 for Mut1 samples, and 2500 for Mut2). Using all these as one system will cause all the peaks to be combined into one consensus peakset. Assume that that the peaks overlap and that there are 2500 peaks in the combined peakset.

My question is whether it is wise to in such a case combine all conditions into one "system", creating a single consensus peakset for them. It is obvious that combining all conditions (and using three contrasts) is much less cumbersome than using three pairwise "systems" (with only two conditions in each). However creating a single consensus peakset for all three will cause the adjustment process of the p-value to correct for 2500 tests for all comparisons. If I would not have combined all, but would compare each pair of conditions separately, then for the WT-Mut1 comparison I would have to correct only for 400 tests performed rather than for 2500.