Mutect multi-sample calling

[davidbenjamin]

That is, multiple samples of the same tumor in space or time, a tumor and a metastasis, a tumor and its cfDNA etc.

First we have to ask collaborators and the community what features they would like and what problems they want to solve.

[samuelklee]

Keep me updated. I think a multi-sample version of ModelSegments would be pretty easy to implement and would hopefully share a similar command-line scheme for specifying allelic-count and denoised-copy-ratio files for the normal/tumors.

Something to think about is that the modeling step probably can be done on a per-sample basis after multi-sample segmentation, and it would be nice to scatter per sample for WGS. There are probably a few ways we can implement this, but let me know if you're planning something similar for M2.

[fpbarthel]

I am very interested in this feature. In my current workflow I am using freebayes (and in an older version simply pileup) to query mutation sites (after doing the actual mutation calling using gATK4 M2) on a cohort level and would be very interested in a GATK naive approach.

P.s. personally more interested in a multi-step mutation calling method than multi-sample calling, where one first calls mutation in multiple samples that are then joined together in a consensus callset, after which the consensus callset is queried individually across the entire cohort of patient resulting in genotype and allele frequencies for each variant across the entire cohort

[ldgauthier]

Since we're using Mutect2 for our mitochondrial variant calling pipeline that's in development, and we want to joint call mitochondria, I'm working on "somatic joint calling". It won't have a joint likelihood model (yet?) the way that germline SNP and indel joint calling does, but it will be able to give you a "squared-off" matrix of calls for each sample at each site that's variant in any sample.

[davidbenjamin]

Closed by #5560.