I am aware that haplotype caller corrects for overlapping reads (e.g. fragment size smaller than sum of PE reads), but I could not find information if MuTect provides this or I should merge/clean PE reads before aligning.
Hmm, good question, I don't know. I'll ask the developer.
@jorgezajgla OK, developer says that if reads overlap at the position under consideration, only one read is counted in the calculation. So you do not need to do any special pre-processing.
What about if the overlapping reads are discordant (i.e. evidence of strand bias or sequencing error)?