Fisher Exact Test, Strand Bias Of Gatk

Question

[juancarlos]

Hi

I have some questions about GATK, I hope to someone can help me :)

What is the Fisher exact test to calculate the Strand Biase in GATK? is weight or no? I haven't see any documentation about the statistical calculation.

Other questions, Is possible to obtain the number of Forwards reads and Reverse reads in a variation position?

Thank You very much

Best REgards

Answer 1

[brentp]

see: http://www.broadinstitute.org/gatk/gatkdocs/org_broadinstitute_sting_gatk_walkers_annotator_FisherStrand.html

It's basically creating a 2x2 contigency table of

1. #ref alleles on + strand
2. #ref alleles on - strand
3. #alt alleles on + strand
4. #alt alleles on - strand

You can always get the read counts using pileup. I'm not sure if GATK reports them.

Answer 2

[Ashutosh Pandey]

Brent is correct. Just to answer your other question: Unified Genotyper doesnt provide the reads for forward and reverse strands individually.

Also, if you have noticed there are two different tags for Strand Bias in GATK output 1) SB 2) FS .

SB is what Brent explained. But usually people use FS that is "Phred-scaled p-value using Fisher's exact test to detect strand bias" .

Normally you remove any SNP with FS > 60.0 and an indel with FS > 200.0

[brentp]

good to know. you mean FS > 60 as a genome-wide cutoff? might want a smaller one for a subset or selected region.

[Ashutosh Pandey]

You may be right. I got these values from GATK web site and have seen few papers using it but I have no idea how they came up with these thresholds. When I use samtools that gives you the p-value for the strand bias, I remove variants with strand bias < .0001 and I assume that translates a Phred score of 40. so a smaller value should make more sense.

Answer 3

[juancarlos]

OK, is very clear the FS formulation and the Thresholds

Thank You very much brentp