plugin trio-dnm2 with deepvariant vcf

[fcliquet]

Hi,

I am trying to move my workflow to this plugin for DNM calls, and noticed something strange in my test dataset. The bcf is produced by calling 3 samples as gVCF using deepvariant before doing a family call using GLnexus.

my command is :
bcftools +trio-dnm2 -P pedigree.tsv data.vcf --with-pAD

produce this line (the samples are the parents followed by the child) :
chr11 70486013 chr11_70486013_C_CG C CG 29 . AF=0.166667;AQ=29 GT:DP:AD:GQ:PL:RNC:DNM:VA:VAF 0/0:30:30,0:48:0,48,959:..:.:.:0 0/0:31:31,0:50:0,120,1439:..:.:.:0 0/1:47:29,18:30:29,0,65:..:-11.7432:1:38

We do know that this variant is true, it was validated with Sanger sequencing.

the DNM score appears awfully low when we have good values supporting the DNM.

• Is it because deepvariant is coding its PL differently?
• if so, is there a way to fix this?
• Should I just skip this and use the naive approach and add criteria on the DP and VAF?

thanks a lot for your help,

Freddy

I attached here the test dataset (and added a .txt for the vcf file for github to accept it):

data.vcf.txt
pedigree.tsv