Differential chip-seq peak calling using macs2

Question

[lucas.michel]

I have been using macs2 for peak calling and posterior differential peak calling. I am working with paired-end data. After calling peaks with macs2 callpeak -B option i use macs2 bdgdiff for differential analysis.

I have observed that, for some contrasts, results look better when I do not use -BAMPE option in the first call to "callpeak". Is that a normal behavior?

When running bdgdiff I have to set the -c (cutoff) at 45 to get reasonable results. Also values in the log ratio of likelihood in the output range from very high values (10³⁵) to very big negative values. Is this the cutoff a reasonable value? are this results normal?

Thank you very much,

Lucas

Answer 1

[Ian]

If you have paired reads then you need to use -f BAMPE, otherwise only the 5' read of each pair will be used and cross-correlation will be used to estimate fragment length, as opposed to using the actual fragment lengths between pairs. Have you compared the 'd' value for each analysis?

I haven't used bdgdiff, in favour of MANORM, diffReps, or DiffBind. diffReps can do without replicates, but diffReps cannot.