Expose VA API via GraphQL

[mattsolo1]

No description provided.

[mattsolo1]

Schema-related links:

The revised schema (YAML format) (diff)
The revised example (YAML format) (diff)
External validator (JSON format)

[phildarnowsky-broad]

thanks @mattsolo1 good idea to put the reference links in here

[mattsolo1]

Related info on the VRS REST API: https://github.com/broadinstitute/gnomad-browser-team/issues/43

[mattsolo1]

The gnomad_methods code provides a clue on what kind of GraphQL arguments we may want to expose to users:

https://github.com/broadinstitute/gnomad_methods/blob/514f634bd4cf31dcdb80a302ac7ba2caeeefffb0/gnomad/resources/grch38/gnomad.py#L530

For now we don't need to support interval queries. Just single variants.

I think the main GraphQL arguments users will be concerned with are by_ancestry_group and by_sex, so we'll want to expose these. They may also want to be able to specify which specific ancestry groups to return (as opposed to getting all of them).

[phildarnowsky-broad]

One small note on this, since the focus is really on VA data more than VRS, I'm going to start calling this the "VA API" for better clarity.

[phildarnowsky-broad]

@mattsolo1 am I missing something, or does the REST API not have a way to specify by_ancestry_group and by_sex parameters?

[phildarnowsky-broad]

@mattsolo1 also the data returned by the API doesn't conform to the schema: in particular, it's missing the type fields, which we're going to need on the GraphQL end.

[phildarnowsky-broad]

@mattsolo1 regarding the type fields, I did think of a way to work around not having those in the response from the REST API if that's not possible, but it would make things much simpler on this end if we did have them.

[mattsolo1]

does the REST API not have a way to specify by_ancestry_group and by_sex parameters?

Not yet.. I still need to add this (should be straightforward).

in particular, it's missing the type fields, which we're going to need on the GraphQL end

Didn't realize this was missing; the REST API is passing along everything I'm getting from the hail lookup, so it must not be returned from this. I'll look into why this is.

[phildarnowsky-broad]

@mattsolo1 looking more closely, it may only be the top-level VRS object that's missing its type, but generally speaking we'll need that anywhere in the data where we have a field that can have one of multiple types--that type from the incoming data will eventually determine the __typename that GraphQL will put in its responses so that clients will in turn know what concrete type is being used.

[phildarnowsky-broad]

Upstream APIs are returning bad data, Dan M. in process of fixing.