I am preprocessing some bam files for differential binding analysis. I have read that some people recommend removing blacklist regions (human) before to do the peak calling. Do you have any suggestions how directly extract the read counts in these regions using maybe samtools?
Keep in mind - when you perform a cross strand correlation analysis to assess peak quality in your data, if you do not exclude blacklisted reads prior to this analysis (which generally uses aligned reads or bed converted intervals from these reads), then you will include these regions in your analysis which may skew the results one way or the other.
Devon is right - the time savings will be nice to call peaks without this step and peak calling shouldn't really be affected much by these regions, depending on which peak caller you decide to use.
Just have a look at a few of the blacklisted regions after your alignment to see what kind of signal you're getting there, and then choose whether to use filtered alignment files or not for whichever downstream analyses (following Sukhdeep Singh's suggestion).
Normally you don't need to modify or do anything with the alignments. After calling peaks, exclude regions that overlap a blacklisted region (use bedtools).
Relatedly, deepTools has an option to ignore reads/signal in blacklisted regions as of version 2.2 I think. This won't do what you need in this particular case, but for other ChIP-seq related things this is useful.
Use intersectBed with