missing node bindings to query graph, in some Automat KP responses?

[saramsey]

Consider the following TRAPI one-hop query graph in JSON format,

{
  "edges": {
    "e1": {
      "predicates": [
        "biolink:related_to"
      ],
      "subject": "n0",
      "object": "n1"
    }
  },
  "nodes": {
    "n0": {
      "ids": [
        "MONDO:0001384"
      ],
      "categories": [
        "biolink:Disease"
      ],
      "is_set": false,
      "name": "myopia"
    },
    "n1": {
      "categories": [
        "biolink:Gene"
      ],
      "is_set": false
    }
  }
}

which I've saved locally as file mwe5.json. When I POST that query graph to Automat-monarchinitiative's production endpoint,

curl -X POST -H 'Content-Type: application/json' -d @mwe5.json https://automat.transltr.io/monarch-kg/query | jq '.' > monarch.json

the JSON response that comes back from the endpoint seems to be missing node bindings (i.e., node bindings to the query graph) for some nodes, like MONDO:0014411, that are referenced in edges in the response's knowledge graph. For example, in this edge (JSON excerpt follows):

      "edges": {
        "uuid:63243a1a-400a-11ef-89e7-6fe0be41fbbf": {
          "subject": "HGNC:20502",
          "predicate": "biolink:causes",
          "object": "MONDO:0014411",
          "sources": [
            {
              "resource_id": "infores:omim",
              "resource_role": "primary_knowledge_source"
            },
            {
              "resource_id": "infores:monarchinitiative",
              "resource_role": "aggregator_knowledge_source",
              "upstream_resource_ids": [
                "infores:omim"
              ]
            },
            {
              "resource_id": "infores:medgen",
              "resource_role": "aggregator_knowledge_source",
              "upstream_resource_ids": [
                "infores:monarchinitiative"
              ]
            },
            {
              "resource_id": "infores:automat-monarchinitiative",
              "resource_role": "aggregator_knowledge_source",
              "upstream_resource_ids": [
                "infores:medgen"
              ]
            }
          ],

So as you can see in that edge, MONDO:0014411 is referenced as the object of the edge, but it is not bound to any node in the query graph, per the list of bound nodes in the same response (excerpted below):

It looks like this node (MONDO:0014411) is in the response KG, and is referenced in edges, but it is not bound to any query node. Is that allowed per the TRAPI spec? Is it allowed because MONDO:0014411 is referenced in an auxiliary graph, namely, the auxiliary graph with these two edges?

[edeutsch]

I am not able to find documentation on what the proper way to handle this. Can someone else find it? If not, we should decide and document it.

[saramsey]

See this Slack post for a vote taken on this issue, in April 2023:
https://ncatstranslator.slack.com/archives/C013Q5TVC87/p1681851155255359

[EvanDietzMorris]

This is really an issue for Plater, not ORION, but it doesn't look possible to cleanly transfer it across github organizations.

The seemingly missing node binding is indeed due to that node being part of a result derived through subclass reasoning, with the "real" edges included as an aux graph. We used to include node bindings for those subclass nodes, with a "query_id" field specifying which query node they corresponded to, but changed the formatting based on complaints/discussions that it wasn't valid. That is to say, this representation is intentional. The linked slack comment/vote seems to indicate people preferred this representation. Whether or not this is how we should do it moving forward is another discussion.

[saramsey]

Thank you Evan! Your reply is super helpful. Sorry I filed this as an ORION issue (as it really is an ARAX issue). I'm going to close this issue.

[saramsey]

Confirmed, this was an ARAX issue, and we have fixed it with a recent patch: RTXteam/RTX#2455