Are MACS2's summit or DiffBind's summit options are recommended for histone marks like H3K4me3, H3K27me3, H3K27ac and H3K4me1?

Question

[Researcher]

Hi All I have used MACS2 for the peak calling with no summit at the default settings and then used Diffbind to call differential peak calls. I am worried as I could not identify any differential peaks around the TSS but they lie somewhere else on the gene bodies. Is it is because I didn't use summit option while calling peaks using macs2 or differential peaks using DiffBind? Will the peak calling with summit option help or improve my calls? Any helpful suggestion is welcome.

Thanks

[colin.kern]

The Macs2 call summit option won't change the boundaries of the peaks called. What it does is add multiple lines in the bed file for some peaks, which will have the same start and end coordinate but different values for the last 4 columns, which are the signal, p-value, q-value, and peak summit. This lets you detect multiple binding events within a peak region which I think is more useful for transcription factor ChIP-seq than histone modifications.

[Researcher]

Hi @colin, Thanks for explaining it well. But then, what do you think about the DiffBind's summit option, will that help? Hi @Rory, Do you have any suggestions on this?

Answer 1

[Rory Stark]

Yes, I would use the summits option in DiffBind to analyze these data.

This is straightforward for the H3K4me3 mark, which is generally narrow. For the other marks, which tend to be enriched over a wider intervals, generally the important thing is to accurately detect changes in enrichment in the vicinity of a certain genomics features (enhancer, TSS, etc). The summits parameter works by finding a point of optimal enrichment among all the samples where enrichment was identified, then creating a consistent window around that point which is used to detect differential enrichment. It is usually more important that the window being tested represents a representative region truly enriched in at least one sample group, than to represent the entire enriched region. Even if it is a subset of the full enriched region, if it is changing significantly between sample groups, that tells you what you need to know about enrichment in/near that feature. If you try to cover the entire region of enrichment, you will include more "background" bases that are not truly enriched, and this noise dilutes the confidence statistics.

Ultimately it depends on exactly what your experiment is designed for; there are some cases where knowing the precise boundaries of enrichment are more important, in which case you may lose some information by re-centering the consensus peaks.

[Aspire]

Ultimately it depends on exactly what your experiment is designed for; there are some cases where knowing the precise boundaries of enrichment are more important, in which case you may lose some information by re-centering the consensus peaks.

Could you perhaps give explain in what type of experiment it would be important to know the precise boundaries of enrichment?

[Rory Stark]

Perhaps when you are looking for fine details of how motifs change at precise binding sites, but generally motif analysis tools can work with peak sets that are not actually centered on the main motif.