Filtering multi-allelic sites in VCF files

Question

[ozankiratli]

I am trying to filter multi-allelic sites in a VCF file to find the true multi allelic sites. Some of these have 1 read in 200, that I want to get rid of. The other alternative looks good. However, when I try to filter the alleles lower than frequency (0.05) with vcftools it gets rid of all variant not just the minor allele. Is there a way to filter out the minor alternate according to the frequency but not the variant completely?

[Ram]

Check out bcftools view -i/-e expressions. The boolean operators might be helpful in navigating your niche requirements.

[ozankiratli]

I tried but both bcftools and vcffilter, they put a cutoff value, then exclude the site all together, I want to keep the site and get rid of the extra allele. This is for poolseq analysis, that's why the read depths matter.

[Ram]

If it is an uncommon operation, you might need to do some manipulation with awk. You could also look for bcftools plugins, but I'm not sure if there's some sort of a library of plugins you could search.

Answer 1

[ozankiratli]

Found the answer!

1. Convert multiallelic to biallelic vcf first

    bcftools norm -m - file.vcf  > biallellic.vcf
   

2. Filter the alternative alleles under certain value

    bcftools view -e "FORMAT/AD[:1]<2 && INFO/AD[1]<5" biallelic.vcf > biallelic-filtered.vcf
   

3. Convert biallelic vcf to multiallelic vcf

    bcftools norm -m + biallelic-filtered.vcf > multiallellic-filtered.vcf
   

[chrchang523]

Note that this effectively replaces all observations of rare alleles with observations of reference alleles, when it's usually more appropriate to replace the affected genotypes with "./.".

[ozankiratli]

That's correct, but depends on the application! I use this for poolseq. So I really don't care about the genotypes. All I need is the allele frequencies. And for my data, I'm pretty confident that those are sequencing errors. I already remove anything below 5%. This was another step to do it.

[chrchang523]

You almost certainly should be excluding the affected samples from your allele frequency denominators. Your approach does not do that.

[ozankiratli]

Can you explain why?
The filtering step can be different for different people, in my case I have 5 samples, and I am excluding the if an alternative allele is lower than 2 in any sample and the count of it is lower than 5 in all samples. Also the rest of my filtering is not included here.

[chrchang523]

Ok, sorry, I should have looked more carefully at the rest of the post. If these thrown-out ALT alleles really are just sequencing errors, this approach is ok. My comments were directed at the case where the ALT alleles were rare but not errors.