Test-drive the GATK tools and Best Practices pipelines on Terra
Check out this blog post to learn how you can get started with GATK and try out the pipelines in preconfigured workspaces (with a user-friendly interface!) without having to install anything.
Incorrect AD values in HC-called vcf and combined gvcf
I am using GATK v3.2.2 following the recommended practices (...HC -> CombineGVCFs -> GenotypeGVCFs ...) and while looking through suspicious variants I came across a few hetz with AD=X,0. Tracing them back I found two inconsistencies (bugs?);
1) Reordering of genotypes when combining gvcfs while the AD values are kept intact, which leads to an erronous AD for a heterozygous call. Also, I find it hard to understand why the 1bp insertion is emitted in the gvcf - there is no reads supporting it:
single sample gvcf
1 26707944 . A AG,G,<NON_REF> 903.73 . [INFO] GT:AD:DP:GQ:PL:SB 0/2:66,0,36,0:102:99:1057,1039,4115,0,2052,1856,941,3051,1925,2847:51,15,27,9
1 26707944 . A G,AG,<NON_REF> . . [INFO] GT:AD:DP:MIN_DP:PL:SB [other_samples] ./.:66,0,36,0:102:.:1057,0,1856,1039,2052,4115,941,1925,3051,2847:51,15,27,9 [other_samples]
1 26707944 . A G 3169.63 . [INFO] [other_samples] 0/1:66,0:102:99:1057,0,1856 [other_samples]
2) Incorrect AD is taken while genotyping gvcf files:
- single sample gvcf:
1 1247185 rs142783360 AG A,<NON_REF> 577.73 . [INFO] GT:AD:DP:GQ:PL:SB 0/1:13,20,0:33:99:615,0,361,654,421,1075:7,6,17,3
1 1247185 rs142783360 AG A,<NON_REF> . . [INFO] [other_samples] ./.:13,20,0:33:.:615,0,361,654,421,1075:7,6,17,3 [other_samples]
1 1247185 . AG A 569.95 . [INFO] [other_samples] 0/1:13,0:33:99:615,0,361 [other_samples]
I have found multiple such cases here, and no errors nor warnings in the logs. I checked also with calls that I had done before on these samples, but in a smaller batch. There the AD values were correct, but there were plenty of other hetz with AD=X,0... I haven't looked closer into those.
Are these bugs that have been fixed in 3.3? Or maybe my brain is not working properly today and I miss sth obvious?