Why might ancestral allele states in 1000G be wrong?

Question

[hyanwong]

According to ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/pilot_data/technical/reference/ancestral_alignments/README ancestral alleles in the human SNPs in 1000 genomes data are determined by comparison with chimp, orangutan, and macaque. Here's an example from the vcf for chromosome 1 (e.g. from http://hgdownload.cse.ucsc.edu/gbdb/hg19/1000Genomes/phase3/):

1       527169  rs563246443     A       G       100     PASS    AC=4;AF=0.000798722;AN=5008;NS=2504;DP=10410;EAS_AF=0;AMR_AF=0;AFR_AF=0.003;EUR_AF=0;SAS_AF=0;AA=g

Which says that the ancestral allele (AA) is "g". But when I look at the alignments in Ensembl (e.g. https://www.ensembl.org/Homo_sapiens/Variation/Compara_Alignments?align=1098&db=core&r=1%3A591289-592289&v=rs563246443&vdb=variation&vf=95730370), I find that the other primate species all have "A" at that locus:

rs563246443 SNP

Human › chromosome:GRCh38:1:591779:591799:1 Chimpanzee › chromosome:Pan_tro_3.0:17:83132247:83132267:1

                     R          
Human      ATCATAGTTGACAATTGCCTA
Chimpanzee ATCATAGTTGACAGTTGCCTA

Human › chromosome:GRCh38:1:591779:591799:1 Orangutan › chromosome:PPYG2:1:229887820:229887840:1

                    R          
Human     ATCATAGTTGACAATTGCCTA
Orangutan ATCATAGTTGACAATTGTCTA

Human › chromosome:GRCh38:1:591779:591799:1 Macaque › chromosome:Mmul_8.0.1:16:77192000:77192020:-1

                  R          
Human   ATCATAGTTGACAATTGCCTA
Macaque CTCATAGTTGACAGTTGTCTA

What gives? Does anyone know why this might have gone wrong in 1000G, and how general the problem might be?

[Kevin Blighe]

I'm not sure. Looking at the data, I'd suggest that the ancestral allele is indeed A. The G variant is a rare allele and is only present in the African 1000 Genomes population, as judged by the dbSNP record: https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=563246443

*in fact, the dbSNP record does not even list the ancestral allele

[hyanwong]

Yes, although from what I’ve read, dbSNP only uses the chimp sequence as the ancestral state, which is much less sophisticated than the 1000G method. I wondered if either the alignments for this region with other species have improved since the 100G calculation, or if there’s a bug in the 1000G AA estimation pipeline

[Jie Ping]

It makes me confused. And, how could we annotate the right ancestral allele for vcf file?

Answer 1

[darink]

Look instead at the EPO multi species primate alignment (that is what 1000 Genomes uses for ancestral calls). There's a "G" there (sorry, Ensemble is currently having problems so cannot share the link).

This muti-species alignment is now quite dated so it's possible that the lower confidence (ie. lower case letter) ancestral allele calls in 1000 Genomes are incorrect. I would trust the current primate assemblies more than the EPO data.