Dear Rory,

I am currently using DiffBind to detect differential chromatin accessibility using ATAC-seq data. We are testing a few different ways of doing this analysis and I have a question on the behaviour of dba.count for one of the options.

This option is to create non-overlapping sliding windows of 300bp at the location of the consensus peak set. The issue is that we have relatively long reads (150bp) so many reads are expected to overlap with 2 adjacent windows.

My question is: how is dba.count handling reads overlapping with multiple peaks? Are they counted for each peak? Discarded because they map to multiple peaks? Or maybe the fraction of read length mapping to each peak is counted (e.g: 80% peak 1 and 20% peak 2)?

I unfortunately could not find information on this in the manual or online. I apologise if I missed this information somewhere.

Thanks very much for your answer!

Best regards,

Julien

Hi,

There are two distinct bits of counting code.  The original code (which will probably be removed when we get to it) counts reads if they overlap the interval by more than half the length of the read (or the interval if it's shorter than the read).

The other code uses "summarizeOverlaps" from the GenomicAlignments package and can be configured by the user, as per the "dba.count" manual page.  (Look for the "bUseSummarizeOverlaps" parameter to dba.count.)

Cheers,

 - Gord