Hello, I am analyzing H3K27Ac ChIP seq samples from paired samples from disease and healthy tissues (each pair from one individual).
Peaks were called with MACS2. The number of peaks and the FRiP (fraction of peaks in reads) is very different between the two experimental groups: IPsample NumberPeaks FRiP Disease1 73186 11 Disease2 56963 6.3 Disease3 106123 18 Disease4 58564 8.3 Normal1 18400 2 Normal2 25585 2.9 Normal3 37038 3.6 Normal4 24459 2.8

I don't know whether the difference is due to real biology, or due to some technical problem. I used DiffBind for detecting differential H3K27Ac. As expected I get much more peaks which are up at the disease tissue. Are there any suggestions for what would be the best normalization in DiffBind in such a case, or how can I try understanding whether this is an artifact? Thank you

If you have a prior expectation of a unidirectional shift in signal intensity toward one sample group (the disease condition), you should avoid normalizing use an RNA-seq method that assumes that there is a core of sites that don't change, and that changes will occur in both directions.

The default methods used by DiffBind should work, namely method=DBA_DESEQ2 and bFullLibrarySize=TRUE. Thhis will result in normalizing only for sequencing depth (total reads) for each sample.

The latest release of DiffBind includes a new interface function, dba.normalize(), that allows for much more flexible normalization. The updated vignette includes a detailed exploration of the impact of different normalization schemes, including the scenario you discuss.