GATK unifiedGenotyper mystery

[matan8]

Hi all,
I am using the GATK to find mutation in s. cerevisiae (SK1 strain). When I am using the unifiedGenotyper in EMIT_ALL_SITES mode, in some expriments I get the in a specific position, there is a variant with very high quality (line from the vcf file):
SK1.chr01 16513 . C T 1765.79 . AC=1;AF=0.50;AN=2;BaseQRankSum=-0.703;DP=75;Dels=0.09;FS=30.976;HRun=0;HaplotypeScore=142.7418;MQ=38.32;MQ0=0;MQRankSum=0.659;QD=23.54;ReadPosRankSum=2.022;SB=-509.39 GT:ADP:GQ:PL 0/1:6,62:68:4.30:1795,0,4

but when I use the unifiedGenotyper in the default mode (EMIT_VARIANTS_ONLY, I think) this variant doesn't appear. Can anyone see the reason why?

[brofallon]

Low quality?

Well, this doesn't really answer your question, but there's a decent chance that variant isn't real. It's very strand-biased (FS is over 30), and the variant allele depth is very low (only 6 of 68 reads have the variant base). Perhaps there's some filtering being done? Perhaps the variant quality is below whatever threshold you're using?

[matan8]

I think you have a mistake about the AD section.
##FORMAT=<ID=AD,Number=.,Type=Integer,Description="Allelic depths for the ref and alt alleles in the order listed">
first ref allels (6 reads) and then the alt allels (62 reads).
As much as I know, all the reads are from the + strand, so strand bias is reasonable. Do I miss something else?

[vivek_]

Just a guess but did you check if the variant is within in a predicted Indel region? Otherwise it does seem strange since --EMIT_ALL_SITES should pretty much give the same result as the normal mode but with homozygous sites included.

[brofallon]

Oops - indeed that's 62 alternate and 6 reference, not the other way, my mistake.
Nonetheless, FS is still very high - that's a phred-scaled p-value so it certainly looks significant. But FS doesn't in itself influence the variant quality score, so even large values wouldn't result in it not showing up. The mystery continues!