PoN for somatic SNVs/indels?

[kpjonsson]

We usually filter variants against a panel of normal samples. This can be done post variant calling by genotyping detected variants in this panel and call anything at a certain threshold in a minimum number of samples. This solution will not work in this framework. I think, rather, that we do variant calling in a normal panel (we have this) to generate a static PoN VCF, similar to GATK's best practices for somatic variant calling.

[evanbiederstedt]

Arguably related to #101

[kpjonsson]

Absent the actual panel, in the interim I'm composing a PoN for the 100 test exomes using the normal samples in the cohort by calling variants in two ways:

• MuTect2 tumor-only mode, where the normal is the input
• Strelka2 germline calling (underway, actually not in this one)
  Then I take unfiltered calls and aggregate in a VCF.
  The stringency of this filter is determined by two cutoffs:
• At what read count is a variant present in the normal sample
• ...and at what PoN count is the variant a FP.
  This is ultimately best decided when a nice and clean PoN exists.

Initial implementation in #228.

[kpjonsson]

I've provided documentation on how to create a PoN for somatic SNVs/indels for exomes:
https://github.com/mskcc/vaporware/wiki/Panel-of-Normals.

I think this can be referred to Trinity for now. It's a CCS task to properly set up a process by which a good panel of normals can be created.