This site is now read-only. You can find our new documentation site and support forum for posting questions here.
Be sure to read our welcome blog!
VariantRecalibrator SNP and INDEL failure rate
I've been struggling with some issues we have been having with the VariantRecalibrator. Here's the story.
We run VariantRecalibrator on our new sample in combination with the previous ones (gVCF file). Few steps: we merge, we run SNPs and then IDELS and finally select regions of interest.
When WES is run, there seems to be no that many problems BUT for panels we get a lot of SNP and INDEL recalibration fails, resulting in FILTER status "." in the final .vcf files.
I've checked if the number of references (so samples other than the new one) affects the success rate, and it does not much. Still on average 50% of our SNP recalibration fail, no matter if i use 10 or 90 samples.
I've been reading the fora and I've encountered a discussion between @Geraldine_VdAuwera and @Wixaros at the thread: http://gatkforums.broadinstitute.org/firecloud/discussion/3893/calling-variants-on-cohorts-of-samples-using-the-haplotypecaller-in-gvcf-mode. Here @Geraldine_VdAuwera mentioned that a low number of sites in a panel may result in an unstable model.
So my question is: Is this just something we have to accept, that for small panels the FILTER field is not much of use for us, since the VariantRecalibrator will fail? Or is there something that we can still do? I am asking, since we need to know how much can we trust the FILTER field and do we need to keep good track of the SNP/INDEL success or failure.
Regards and sorry for the wall of text.