gnomAD minor allele frequencies for protein variants

[jstone-dev]

MaveDB obtains gnomAD minor allele frequencies for variants assayed at the DNA level, but not for protein variants. Its data model allows each mapped variant to be linked to several gnomAD variants.

Should we provide minor allele frequencies for protein variants? This would involve reverse translation, and we could use our existing variant_translations table for this purpose, since gnomAD linkage is already limited to variants with ClinGen IDs. In the UI and perhaps in API responses, we would presumably give the sum of the allele counts and frequencies, and we would omit any ancestry-specific information, perhaps excepting the case where there is agreement in faf95_max_ancestry.

I have needed to do this to annotate a protein variant set, and I have code ready in case we want to integrate it into MaveDB. But it seems to warrant some discussion about (a) whether it's appropriate and (b) how to make our approach clear to users.

[afrubin]

I strongly agree that we should show this information. This does seem related to other discussions we've had around adopting the GA4GH Cat-VRS for associating nucleotide variants to protein variants but we can implement this without adopting that standard.

One way to make this clear to users would be to label this as something like "maximum gnomAD minor allele frequency" or similar and providing the highest allele frequency for any of the nucleotide variants that can produce the given amino acid variant. I'm assuming that this is what most users are interested in.