Union Peaks For Peaks Called On Different Strains

Question

[Sander Timmer]

For a bunch of strains we have Chip-Seq on which we have been calling peaks using SWEMBL. We now want to generate an union peak set and I was wondering if people can shed some insight into this and things I should consider.

Is there a "standardised" way of doing this?

If not, when do you merge peaks together (how many bp overlap) for example if strain 1 has a peak from 1 -20 and strain 2 from 18-30 should we merge this (I think we should) but then how do we make sure we don't get greedy and end up with some massive wide peaks (we have 50+ strains).

Any insights or useful tips/tools/scripts people can recommend for doing this?

Answer 1

[Sukhi Singh]

Why you want to generate a Union Peak set. I think the better way to do would be extend you reads strand specifically to the initial fragment size which was recovered during sonication (in our case its 250 bases). After extension of the reads in all the strains, you would do overlay the peaks can check for a specific threshold (1bp to complete overlap), how many of them overlap. This is a measure of peak union and you can get the output in a way, how you want like, genomic area or the number of reads overlapped in all or a specific strain or the complete area of read/strain 1 overlapped by another etc. depending upon what you want.

If this suits you, then the tool used for this intersections could be an R package like chipPeakAnno or using intersectBed of bedtools

Merging can also be performed using mergeBed of bedtools, in my case when I was merging peaks, it introduces bias, as when you merge the peaks, the peak height is added as well which accounts for the false enrichment in the merged area.

Sukhdeep