Greetings everyone,
I am tearing out my hair trying to incorporate genotype likelihoods into several population genetics measures. I asked a question on the GATK forums, but didn't get a useful response (in terms I could understand). To be honest I was put off by the help I received.
http://gatkforums.broadinstitute.org/discussion/2923/re-scaling-genotype-likelihoods
I thought I would try my question here.
The genotype likelihoods (VCF field: PL ) are altered so that the most likely genotype has a probability of one or a Phred score of zero. This is causing me trouble integrating over the other possible genotypes. Alternatively I thought I could use genotype quality (GQ) for the calculations, but it throws everything off.
How are people dealing with this? Why does Unified Genotyper set the most likely genotype to a Phred Score of Zero? How can I account for the uncertainty of best genotype if it always has a probability of 1?
Any advice would be great! Thanks everyone.
example:
GT:AD:DP:GQ:PL
1/1:0,3:3:6:57,6,0
•
written
by
Zev.Kronenberg
According the the GATK team itself currently you can not back-calculate non-normalized likelihoods from UnifiedGenotyper (see this question on their message boards: calculating-non-normalized-genotype-likelihoods)
They do suggest you can go to GitHub and alter the code yourself to possibly output the values you want.
How does using the GQ throw everything off? That does seem like the best way, given the output, to account for the uncertainty. If you haven't read it the GATK's documentation on their VCF output may be of some help: How Should I Interpret VCF Files Produced by the GATK
So this is the equation I am playing with. I tried using GQ in the bottom derivation, not summing across the other genotypes. Didn't work
http://genome.sph.umich.edu/wiki/Genotype_Likelihood_based_Inbreeding_Coefficient
Log in to answer this question.
