Missing variant after HaplotypeCaller joint calling

[drmjc]

Hi all,
A while ago I generated a single-sample VCF via HaplotypeCaller (v2.5 or v2.8), and we found a great candidate variant, with lots of read depth support. Now after some time i've redone this analysis with HC v3.1, now using joint calling, and the variant has dissapeared. To rule out the impact of HC v3.1, i called the single sample with HC v3.1 and the variant is again found.

I've attached GenomeBrowse screenshot, where the bottom 2 tracks show the strong read depth supporting the variant, and that the variant does exist in the single-sample VCF. the top track shows that the variant is missing from the multi-called VCF.

given that the new HaplotypeCaller "is so sensitive that it cries at the movies" (Broad's words not mine), i'm perplexed why such a strong variant didn't make it through the multicalling.

Has anyone seen similar, or better yet have an explanation?

cheers,
mark

[drmjc]

bump, has anyone also seen this?

[bruce01]

Haven't seen this but my understanding is that the HC3.1 is meant to be used to make sample specific GVCFs which are then joint genotyped using GenotypeGVCFs. If you are giving multiple input to HC3.1 then you may get the '6+ alleles' error at your variant site and so it is skipped(?) Could make a small sample file to test this.

Also you seem to have done VQSR on the top joint VCF (from the filename), if the variant is novel (?) could be an issue with it not getting through the VQSR. You can add the variant to a 'safe list' to see if that helps.

Good luck.

[drmjc]

Thanks bruce01. Using HC v3.2 properly (ie via individual gVCF) does indeed recover the variant. For the public record, it wasn't in the raw VCF file, so not a VQSR problem

cheers,
Mark