some genotype calls are wrong, why?

jimmikeself

Hello the team,

For some genotypes, it seems are wrong, I know it's model based, and base q, map q, etc are considered in the model. I also read this link: http://gatkforums.broadinstitute.org/discussion/1235/why-didnt-the-unified-genotyper-call-my-snp-i-can-see-it-right-there-in-igv#latest But my case are special, the format is (ref allele count)/(alternative allele count) genotype call: 22/24 0/0 109/125 0/0 85/109 0/0 26/32 0/0 40/161 0/0 195/6 1/1 239/5 1/1 83/6 1/1 46/28 1/1

In one case, the two variants are adjacent to each other. In some case, they are one base indels.

Thanks,

Jim

Geraldine_VdAuwera

Hi Jim,

Can you give some more supporting info? For example, some IGV or pileup screenshots that show the reads and base calls at those loci? Without seeing more it's impossible to interpret the genotype numbers.

jimmikeself

Hi Geraldine, Thanks. Here is the genotype call and attached is the IGV snapshot

chr10:124329672 T T/T chr10:124329673 C C/C

I can email you a small slice of the BAM file, since it has some meta info, I do not want to post it here.

Thank you, -Jim

ebanks

What is your command-line for calling? Are you possibly running with the generalized ploidy model and asking to genotype a ploidy of 1?

jimmikeself

Eric, Thank you. I am using genotype_likelihoods_model, here is the cmd: java -Xmx4g -jar GATK
-T UnifiedGenotyper -nt 8 -R HG19REF -I test.bam -o test.vcf -metrics OUTPUT/test.txt -dcov 250
-L chr10 -glm BOTH -stand_call_conf 30 -stand_emit_conf 30 -baq CALCULATE_AS_NECESSARY
-G Standard -A AlleleBalance -A DepthOfCoverage
-A HomopolymerRun -A QualByDepth

-Jim

jimmikeself

The version is version 1.6-11-g3b2fab9 on Linux 64

ebanks

Hi Jim,

I would try updating to the latest version of the code and turning off BAQ (which has been known to kill multi-nucleotide polymorphisms). If that doesn't work, it would be good to see what the HaplotypeCaller calls at this position.

jimmikeself

Eric, Thanks a lot! That's exactly the cause. Can you give me a clue about the effect on false positive calls if I turn it off? To my understanding (please correct me if I am wrong or incomplete), BAQ was introduced to reduce false positive calls around alignment induced indels. I wonder how much benefits it has. -Jim

ebanks

BAQ has tons of benefit for regular SNPs, but it tends to kill more complex events like MNPs (as you saw firsthand) and indels. This is precisely why we have developed the HaplotypeCaller as the successor to the UnifiedGenotyper.

jimmikeself

Eric, Thanks for the insights. So HaplotypeCaller kept the benefits and eliminated the drawbacks on complex events? Is 1000genome project using HaplotypeCaller? I'd like to know more about it. Could you point me to some docs or presentations about it, or the comparison of it and UnifiedGenotyper. What is the major difference? Thanks, -Jim

Geraldine_VdAuwera

Hi Jim, have a look at this presentation:

http://gatkforums.broadinstitute.org/discussion/1576/slides-from-the-broad-institute-medical-and-population-genetics-program-meeting-9612

jimmikeself

thanks!