SNV calls data release 2

[fpbarthel]

To-do's related to SNV calls for the second data release

• [x] Currently we are using one PON for the entire cohort as per https://github.com/TheJacksonLaboratory/GLASS/issues/46. Consider moving to cohort-specific PONs, which has worked well for CNV see https://github.com/TheJacksonLaboratory/GLASS/issues/45. For new "per-batch" PON specification see https://github.com/TheJacksonLaboratory/GLASS/issues/101. Implemented in the current version of the pipeline
• [x] SNV calls on a set of samples seems to be "missing" (or "disappeared"?) from the database, this needs further investigation and to be addressed post submission. See https://github.com/TheJacksonLaboratory/GLASS/issues/105. No longer relevant since we called again from scratch
• [x] Add filtering information from M2 to database, since its possible variants are filtered in one instance (eg. primary) or a subject's tumor only. Done
• [x] Parallel calling using bcftools, or M3 if already available. Throw away parallel calls across whole dataset and focus on parallel calls within a patient. Done using GATK 4.1 M2
• [x] Check GATK 4.1 M2 VCF files for completeness/missingness @fsvarn
• [x] Compare GATK 4.0.x and GATK 4.1 using the Australia samples @anderkj2
  • [x] Compare GATK 4.0.x and GATK 4.1 Mutect2 variant calls
  • [x] Compare freebayes and mutect2 read depths
  • [x] Compare freebayes and mutect2 allele frequencies
• [x] Collect coverage statistics per gene and sample. @fpbarthel
• [x] Assess the impact of purity on sensitivity of M2 to detect variants, adjust mutational rates/counts by purity @fpbarthel

[fpbarthel]

@anderkj2 compared GATK 4.1.0.0 Mutect2 (GATK in figure) variant calls versus GATK 4.0.x Mutect2 calls (freebayes in figure since later combined to a multisample vcf using freebayes to get allelic frequencies):

While there is a large overlap of about 20k variants, there is a huge number of GATK 4.0.x M2-only variants (22k). This could be explained by several factors:

• Multi-sample filtering in GATK 4.1 versus single-sample filtering in GATK 4.0
• Additional filters in GATK 4.1

Comparison of read depths between GATK 4.1 M2 and from Freebayes using the 20k variants that were analyzed in both demonstrates a lower depth from Mutect2 compared to Freebayes. Likely this can be explained by the more stringent read filters that apply in M2. The tail of indels that have a higher read depth in Mutect2 are all indels that Freebayes was not able to accurately count reads for.

[fpbarthel]

Closing this issue, see also Slack discussion @fpbarthel and @anderkj2 .

Two open issues moved to new separate issues