Filter multisample vcf for denovo variant

Question

[finswimmer]

Hello,

I have a multisample vcf file (3 samples) and I'd like to get all denovo variants for a specific sample. I tried bcftools:

bcftools view -s SAMPLE_ID -x All-final.vcf

The problem is, that some sites can be multiallelic. So the above command would e.g. not find this line (The third sample is the one I'm interested in):

chr5    38528951    rs762238623 GACAC   GAC,G   1204.93 PASS    .   GT:DP:AD:RO:QR:AO:QA:GL 0/1:10:2,6,0:2:75:6,0:212,0:-15.9529,0,-3.87444,-16.555,-5.68062,-22.6541   0/1:40:10,21,3:10:343:21,3:677,105:-51.4565,0,-21.4531,-45.9266,-19.2432,-72.8601   1/2:39:0,19,10:0:0:19,10:622,279:-72.2747,-22.0435,-16.3239,-50.201,0,-47.1907

Because the requirements are not fullfilled:

-x, --private print sites where only the subset samples carry an non-reference allele. Requires --samples or --samples-file.

So, what's the best way here to find all denovo variants in a given sample?

Thanks.

fin swimmer

Answer 1

[Pierre Lindenbaum]

try GATK SelectVariants https://software.broadinstitute.org/gatk/documentation/tooldocs/current/org_broadinstitute_gatk_tools_walkers_variantutils_SelectVariants.php

Generating a VCF of all the variants that are mendelian violations. The optional argument '-mvq' restricts the selection to sites that have a QUAL score of 50 or more

 java -jar GenomeAnalysisTK.jar \
   -T SelectVariants \
   -R reference.fasta \
   -V input.vcf \
   -ped family.ped \
   -mv -mvq 50 \
   -o violations.vcf

I've also written: http://lindenb.github.io/jvarkit/VCFTrios.html

[finswimmer]

Hello Pierre,

checking for mendelian violation is not exactly what I ask, but for more needs this is also very good.

Thanks a lot.

fin swimmer

[Pierre Lindenbaum]

what's your definition of l 'denovo variants ' without the context of a trio ?

[finswimmer]

Without the context of a trio my definition for denovo is a non-reference allel that only occur in on specific sample compared to other samples in the multisample vcf.

But within the context of the trio your are absolutly right, that every mendelian violation is at least suspicious.

fin swimmer

[Pierre Lindenbaum]

denovo is a non-reference allel that only occur in on specific sample compared to other samples in the multisample vcf.

i would say it's a "rare variant" :-)

[finswimmer]

Ok, if this is the right term :)

Even if my initial problem is solved, I'm still interested in how to filter those rare variants for a given sample within a multisample, multiallelic vcf.

fin swimmer

[Pierre Lindenbaum]

still with GATJ SelectVariants using the option -select someting like '-select "AC<1" see the GATK doc/ JEXL.

Answer 2

[Chadi Saad]

use genmod to annotate your variants with genetic models: