We are running Mutect2 with the option to ignore soft clipped bases. This has significantly reduced our false positives, but the ALT and REF counts from the remaining reads do not seem to reflect what we see in IGV. Moreover, this seems to occur adjacent to the softclipped boundary. Could the softclipping index be off by 1?
Here's counts of alt and ref for one of these variants that we found:
ALT REF IGV-ALT IGV-REF use softcipped 35 239 39 242 ignore softclipped 37 235 4 237