Fisher Strand in GATK, a good measure to detect Strand Bias?

Question

[cristianrohr768]

Hello,

I follow the GATK good practices to call variants on 11 samples from a custom truseq design. I do Joint genotyping and apply Hard Filters to my data http://gatkforums.broadinstitute.org/discussion/2806/howto-apply-hard-filters-to-a-call-set.

In a second test, I skip the joint genotyping step, and just apply hard filters to my individuals VCFs.

With both strategies I found the same variant in the same individual which is a pathogenic variant for a disease I was interested in.

For this variant I have a read depth of 180, and I see A:103 and G:77, G is the reference, there aren't homopolymers near. But the problem is that this position is ONLY covered by forward reads.

The FS calculated by GATK for this position is FS=0.000

FS is Phred-scaled p-value using Fisher's Exact Test to detect strand bias (the variation being seen on only the forward or only the reverse strand) in the reads. More bias is indicative of false positive calls.

So a value of 0 must be indicative of no bias, but this position is only covered by forward reads, so I undoubtedly had strand bias in this position.

So what's going on? I can have confidence on this variant?

Answer 1

[Zaag]

It tests the difference between variant and ref alleles so probably the ref alleles on that position are also on the forward strand, so the bias is the same for ref and var reads so there is no reason there to discard the variant.

Answer 2

[Pablo Marin-Garcia]

One nice explanation could be found in this samtools mailing list post

If you have only reads in one strand you can not "have" strand bias, or at least you can not measure it, since the other strand has 0 reads.

The strand bias measure if the proportion of alleles is statistically different between strands. As you don't have reads in the other strand, you can not measure it. You can only measure allele bias (departure of the 0.5 if you are expecting germline heterozygotes)