Q: impact of sequencing throughput on peak calling

[SplitInf]

Dear @taoliu

Thank you for macs and for the constant improvements you have been adding.

Use case I have some histone chip data on cancer cell lines where they were sequenced extra deep by the sequencing service centre (>80M reads).

Describe the problem Surprisingly, despite all cell lines being quite similar, the number of peaks called varied. A few deeply sequenced samples have significantly more peaks being called. Visually inspecting peak calling indicates that there are noticeably more small/noise-like peaks being called. It's also surprising since the input samples were sequenced deeply and would, in theory, help to mitigate the signal/noise problem.

Describe the solution you tried QC results indicate all samples were successful, with an average RiP% > 15% ; RiBL% <5%. Down-sampling of the samples in question helped to lower the number of peaks being called. My question is: Are these behaviours expected and there is a recommended way to handle samples with uneven sequencing depths? Thanks!