What about the ancestral stat?

[jiangzy26]

Hi, I found tsinfer and relate considered the ancestral stat, but I didn't found it in SINGER, even if I convert the result to ts format, I don't think they are the same thing, could you share how to take the ancestral stat of allele into consideration? Thank you very much.

[moshejasper]

Hi @jiangzy26 . As another user, let me share my understanding (if it gets corrected, let it be corrected).

By 'ancestral stat' I presume you mean 'ancestral allele'.

My examination of the code (especially the 'Sampler' object which reads in the vcf file) suggests that:

1. Singer only uses biallelic sites. It achieves this by:
   (a) first, checking the length of REF & ALT fields (they both have to be <= 1), then
   (b) then, it counts the 1s & 0s across vcf genotypes field to ensure files are genuinely biallelic.

Note that if your data was biallelic but your alt field wasn't (e.g. you had a 'G' as possible alternative site that wasn't present in your data, e.g. because you subsampled) this site would be excluded. You should always use something like bcftools to re-calibrate the vcf before using this.

2. The '1' value (i.e. the ALT allele) is treated as the ancestral allele within SINGER (mutations are only added at a site in code if genotype=='1'.

CONCLUSION:
If you want SINGER to take into account your ancestral alleles, you need to rewrite your VCF file so that the REF column corresponds exactly to the ancestral alleles you are using.