rare variant missing in the vcf file

[crang]

Hi all,

I'm new to next-gen, so I'm sorry if my question sounds lame.

I have genotype data for large group of samples (all from one population and same ethnicity) in vcf format. For some samples I see only AA and AB genotypes, but not the rare BB, and for the rest of the samples i have AB. My question is: is it possible the BB variant to be missing because of filtering somewhere in the variant calling steps? Or there is another explanation?

[crang]

Maybe I should clarify more: I do not see any BB genotypes for the reference allele in the vcf file, i.e. I have always 1/1 or 0/1, but never 0/0. The variant calls are done with samtools|bcftools

[swbarnes2]

And you have empirical evidence that your vcf should have BB?

Is it possible that the genotype is in a repetitive region, which might make it hard to get accurate genotypes?

For instance, do all the AB genotypes look to be 50/50 splits read-wise? Or is it possible that most are 75/25, with only a few 50/50?

[vivek_]

Originally Posted by crang

Maybe I should clarify more: I do not see any BB genotypes for the reference allele in the vcf file, i.e. I have always 1/1 or 0/1, but never 0/0. The variant calls are done with samtools|bcftools

Well unless you are asking samtools to output the consensus you will not find the 0/0 sites since there is no evidence of a variant allele present.

[ymc]

You should run your bams through IGV. Then you might find the reason for your problem.

[JeremyCarroll]

I think vivek has it. VCF is generally used for recording variant calls, not reference calls. A 0/0 is a reference call, and so may be omitted by some tool in your tool chain. I suspect it is not rare, but actually the most frequent genotype.

[swbarnes2]

If you are making the mpileup with multiple .bams, which is a pretty good idea, you should see plenty of 0/0 calls.

But yes, if you are only putting one .bam at a time in, you will not see 0/0 calls. The file would be enormous if they were all in there.

[crang]

Originally Posted by JeremyCarroll

I think vivek has it. VCF is generally used for recording variant calls, not reference calls. A 0/0 is a reference call, and so may be omitted by some tool in your tool chain. I suspect it is not rare, but actually the most frequent genotype.

Yeah, I checked - its the most common and is a ref call. Its my bad that I stated it wrong in the beginning. So if I have 1/1 and/or 0/1 I'll know which samples are possible 0/0 and fill the gaps. But here comes the question how to check if the sample has been genotyped for that variant or not. If not, it will be missing and not 0/0. First that comes in my mind is to check for any good coverage in those positions. Correct me if I'm wrong.

[swbarnes2]

What you could do is make a vcf at every single point, and then filter it by position so it only shows you the loci of interest.

[vivek_]

Originally Posted by crang

Yeah, I checked - its the most common and is a ref call. Its my bad that I stated it wrong in the beginning. So if I have 1/1 and/or 0/1 I'll know which samples are possible 0/0 and fill the gaps. But here comes the question how to check if the sample has been genotyped for that variant or not. If not, it will be missing and not 0/0. First that comes in my mind is to check for any good coverage in those positions. Correct me if I'm wrong.

If you are genotyping multiple samples you can either

a) Make variants calls together for all sample bams, this is called multi-sample variant calling and actually improves the accuracy of the genotyper

or

b) Call each sample bam separately, collect an intersection set of variant locations across all samples into a bed file and re-genotype all the samples for those regions

[crang]

Thanks a lot for all suggestions. Will see what can be done.