Automat-robokop returns 422 errors for some sub-queries

[colleenXu]

When running Pathfinder example 5 (saved response), we send sub-queries to automat-robokop with 1 ID. These sub-queries all fail with status code 422 and the console logs say "message":"A query was made with the following predicates, but none of them or their descendants are in the graph queried: ['biolink:response_affected_by']".

I see MetaEdges with predicate response_affected_by in automat-robokop's meta_knowledge_graph response, so I think that's why BTE makes these subqueries.

[colleenXu]

We've let the Automat team (Evan Morris) know. Here's the info Evan has given us so far in Slack, bold emphasis is mine:

Thurs 10/24:

1:18 PM Pacific: Hmm no this sounds like a bug, it is supposed to indicate exactly what it says (although I’m second guessing whether 422 is the best way to respond to that, I just didn’t want to respond with a 200 and an empty result when the predicate could never return results), but anything in the meta kg should be valid. I’ll look into this.

1:30PM Pacific: I think this still represents a bug because we want to support either direction.. I’ll fix it

Fri 10/25, 8:28AM Pacific:

so - this is caused by the biolink model not having affects_response_to designated as the inverse of response_affected_by under response_affected_by (though inverse: response_affected_by is specified under affects_response_to). I'm realizing I'm not sure what the rules are for this.
Sierra Moxon (SRI) are inverse: tags supposed be included under both predicates going both ways? only under canonical or non-canonical forms? I think I see instances of both..
Either way I think what BTE is doing is not wrong here - response_affected_by is in the meta knowledge graph for robokop, and we should be able to support queries with it. Though I think this is a can of worms - should we be exposing or using non-canonical predicates in translator at all? I'm having trouble thinking of a use case where it's needed and handling them just makes everything more complicated and slower, shouldn't we just make users use canonical predicates?