BCFtools will not filter fixed sites

[RobinC]

Hello all,

I'm encountering a surprising problem while filtering a vcf file in bcftools (v1.1-52-ga2e5b56). It doesn't look like a proper bug but leaves be bewildered enough.

I'm trying to filter out all variant sites, and retain only fixed sites. So, logically enough, I apply a filter to discard sites where minor allele count is above zero, with :

bcftools view --max-ac 0:minor

It mostly works, but only mostly : at some sites, the odd heterozygous individual is retained (0/1:2,1:3:30:30,0,65 for ex in a mostly 0/0 line). This is not a problem of a deadlock between two alleles with equal frequency (thus no minor), since this odd "1" is the only one with 100+ "0"s.

So, I believe there must be some kind of hidden tolerance threshold somewhere, but, does any one have a workaround..?

Thanks a lot for insight !

Robin

[RobinC]

With a quick edit precision: this also applies to VCFtools..! here's an example :

First, just scanning the VCF file manually:

bash-4.1$ gunzip -c data.vcf.gz | grep '0/0' | grep '0/1' -v | grep '1/1' -v | grep '0/0' -c
2408 ##these are ref:ref homozygous

bash-4.1$ gunzip -c data.vcf.gz | grep '1/1' | grep '0/1' -v | grep '0/0' -v | grep '1/1' -c
46180 ##these are alt:alt homozygous

bash-4.1$ vcftools --gzvcf data.vcf.gz --max-mac 0 --recode --stdout | grep -c '0/1'
0 #that's what I want, but then:

bash-4.1$ vcftools --gzvcf data.vcf.gz --max-mac 0 --recode --stdout | grep -c '0/0'
0 ##none of my 2408 ref:ref homozygotes has been retained !

bash-4.1$ vcftools --gzvcf data.vcf.gz --max-mac 0 --recode --stdout | grep -c '1/1'
46122 ##and some of the ref:ref have been dropped too.

This is surprising, as both BCFtools and VCFtools explicity distinguish the notions of "minor allele" and "non-reference allele". So this should NOT be an issue.

All ideas very much welcome !!

Best

Robin