In the process of analyzing ChIP-Seq replicates with DiffBind (drosophila data, 3 replicates for each sample, choosing DESeq2 as the analysis method), peaks with very small Fold-change were found to be significant.
The median absolute value of the fold change among the significant peaks is 1.55 and the mean is 1.93 (not in log scale). The Concentration is also not that high, the median among the significant values being 225 reads.
I have plotted an MA plot; the red points are the significant ones, the black points are the non-significant ones. I have plotted the non-significant points after plotting the red points, so if a non-significant peak would have a large fold, it would be easily visible.
All points with even slight log fold changes are significant. I wonder, why could this be the case?
For example, the peak which has the smallest fold change, and is still significant is (Fold and Conc are in log2 scale here):
seqnames start end width strand 3L 11657351 11659235 1885 * Conc Conc_Mut Conc_WT Fold p.value FDR 8.48 8.63 8.3 0.23 0.00655 0.0376
Individual counts (not in log2 scale):
S7 S8 S9 S1 S2 S3 404.42 391 396.29 305.76 317.14 321.53
Had these been the results with genes in RNA-Seq, I'd be surprised. Is it merely the fact that the ChIP-Seq has much less "genes" to make a multiple hypothesis correction for? Could it be something else?