Something I've seen with ATAC-seq analysis is the extension of peak summits (ie. "Peak summits were extended +/- 250 bp and filtered for nonoverlapping, maximally significant 500-bp peaks"). Even DiffBind does the extension with the summits parameter in dba.counts.
I was hoping someone had a good explanation as to why extended summit peaks are used instead of the actual peak sizes called by the peak caller (MACS2 in my case). Moreover, what is the justification for choosing a specific extension size. I've seen 250 bp used frequently, but I've also seen smaller extensions of 100 bp.
Lastly, I'm not sure of the best way to extend peaks summits, as in, which tool to use and whether to use the summits file output from MACS2 or the summits position in the narrowPeaks file.