Workflow B problem

[cbizon]

See ranking-agent/aragorn#26

[patrickkwang]

We can take a look at this again after the ICEES KPs are registered at TRAPI 1.2.

[patrickkwang]

This is the query in question:

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids": [
                        "MESH:D056487"
                    ],
                    "categories": [
                        "biolink:DiseaseOrPhenotypicFeature"
                    ]
                },
                "n1": {
                    "categories": [
                        "biolink:DiseaseOrPhenotypicFeature"
                    ]
                }
            },
            "edges": {
                "e01": {
                    "subject": "n0",
                    "object": "n1",
                    "predicates": [
                        "biolink:has_real_world_evidence_of_association_with"
                    ]
                }
            }
        }
    }
}

It works when sent to ICEES DILI directly, but not through Strider.

[patrickkwang]

The identifier that ICEES prefers for DILI is DOID:2044. It seems to accept both the DOID and the MESH just fine. The problem is that the node normalizer knows DILI as a biolink:Disease, and we replace the categories for n0 with ["biolink:Disease"]. ICEES returns 0 results when we do this, regardless of which CURIE we use. Sending both categories appears to work. So there are at least two possible solutions:

1. Strider sending the provided category and/or all relevant biolink categories (all the way up the tree).
2. ICEES handling biolink:Disease here the way we'd like it to.

I think we should definitely do 2, and perhaps 1 also. I will make an ICEES issue. What do folks think about how Strider handles qnode categories? Two specific questions:

• What should we do about categories provided when ids are also present? Currently we ignore them and replace with node-norm results.
• Should we provide upstream categories (e.g. biolink:DiseaseOrPhenotypicFeature for nodes with biolink:Disease)?

[cbizon]

This is a good question. My feeling has been that if we're sending a curie then we should not (need to) send any categories at all. But I think that not all KPs were ok with that.

Fundamentally, this seems to be an ICEES problem, and I don't think sending parent categories should be done...

[patrickkwang]

ExposuresProvider/icees-api#181