Unusual AF data (I think)

Question

[rightmirem]

I'm working with 70 samples of whole genome data (the whole genome of 70 individuals). I'm returning to the field after some time away from my academic training, and I don't have a lot of experience with whole genome data from which to draw comparisons. At this time, I dont have any demographic info about the samples.

I've been provided with BAM files, against which samtools mpileupand bcftools call has been run. Then the BCF files were bcftools merge-ed into one large file of 70 samples.

I'm currently working on validating the SNPs in the BCF files. One interesting data curve I've noticed is the Allele-frequency/number-of-SNPs.

Graphs

So this is the AF on the X, and number of SNPs at this AF on Y. The "jump point" is at AF=0.49...so basically, where the SNPs go from hetero- to homozygous.

I would intuitively expect (in a random population) that this curve would be closer to a bell curve...or at least more randomly distributed. And yet this is quite the opposite.

I'm also curious if the Ti/Tv ratio is significant...as there's a spike at the same place, and I wouldn't expect to see the ratio change amongst SNPs at all (should I?)

[Matteo Schiavinato]

Could you add axes labels also on the plots shown, to help readers?

[trausch]

Did you set missing genotypes to ref (-0) when using bcftools merge?

[Kevin Blighe]

Hello, yes, the spike at 0.49 is unusual for a multi-sample VCF/BCF, but it could be explained if, when you merged your samples, the AF field was not updated by BCFtools. Did your colleagues split multi-allelic sites and left-align indels prior to merging?

I would fully expect, however, a binary distribution for AF (0.5 or 1.0) if I was analysing a single-sample VCF. If you have a single-sample VCF, then your AF should be either 0.5 (heterozygous with AC=1) or 1.0 (homozygous with AC=2). If you then merge this with another sample that does not contain the same variant, then you will get the following:

• AC=1
• AN=4 (2 samples * 2 chromosomes)
• AF=0.25 (AC / AN)

The phenomenon that you see may in addition be related to sample population. You have not mentioned from where they are based. If it's a community that has not mixed (genetically) with other groups of people for geographical, religious, ethnic, or other reason, then I would fully expect a skewed AF distribution.