No category information for node in KG returned from query

[saramsey]

I have a TRAPI query message,

{
  "message": {
    "query_graph": {
      "edges": {
        "e00": {
          "object": "on",
          "predicates": [
            "biolink:affects"
          ],
          "subject": "sn"
        }
      },
      "nodes": {
        "on": {
          "categories": [
            "biolink:Gene",
            "biolink:Protein"
          ],
          "ids": [
            "NCBIGene:4314"
          ],
          "is_set": false
        },
        "sn": {
          "categories": [
            "biolink:ChemicalEntity"
          ],
          "ids": [
            "CHEBI:29103"
          ],
          "is_set": false
        }
      }
    }
  },
  "parameters": {
    "tiers": [
      1
    ]
  },
  "submitter": "infores:arax"
}

in a file message-arax-sent-to-retriever.json, and when I send that message to the development Retriever API via this command,

curl -s -X POST \
       -d @message-arax-sent-to-retriever.json \
       -H "Accept: application/json" \
       -H "Content-type: application/json" \
        https://dev.retriever.biothings.io/query | jq . > retriever-response.json

I see in the response message's knowledge_graph, a node with id of CHEBI:29103, whose categories attribute is an empty list.

        "CHEBI:29103": {
          "categories": [],
          "attributes": []
        }

or shown as a screen-cap here:

This, despite the fact that NodeNorm provides type information for this node; the NodeNorm query

curl -X 'GET' \
  'https://nodenorm.ci.transltr.io/1.5/get_normalized_nodes?curie=CHEBI%3A29103&conflate=false&drug_chemical_conflate=false&description=false&individual_types=false&include_taxa=false' \
  -H 'accept: application/json'

produces the response

{
  "CHEBI:29103": {
    "id": {
      "identifier": "CHEBI:29103",
      "label": "potassium(1+)"
    },
    "equivalent_identifiers": [
      {
        "identifier": "CHEBI:29103",
        "label": "potassium(1+)"
      },
      {
        "identifier": "CHEBI:26219"
      },
      {
        "identifier": "CHEBI:49685"
      },
      {
        "identifier": "CHEBI:8345"
      },
      {
        "identifier": "UNII:295O53K152",
        "label": "POTASSIUM CATION"
      },
      {
        "identifier": "PUBCHEM.COMPOUND:813",
        "label": "Potassium Cation"
      },
      {
        "identifier": "CHEMBL.COMPOUND:CHEMBL1233793",
        "label": "POTASSIUM ION"
      },
      {
        "identifier": "DRUGBANK:DB01345",
        "label": "Potassium cation"
      },
      {
        "identifier": "MESH:C068616",
        "label": "liver regeneration factor 1"
      },
      {
        "identifier": "CAS:125352-08-1"
      },
      {
        "identifier": "CAS:138363-30-1"
      },
      {
        "identifier": "CAS:24203-36-9"
      },
      {
        "identifier": "HMDB:HMDB0000586",
        "label": "Potassium"
      },
      {
        "identifier": "KEGG.COMPOUND:C00238",
        "label": "Potassium cation"
      },
      {
        "identifier": "INCHIKEY:NPYPAHLBTDXSSS-UHFFFAOYSA-N"
      },
      {
        "identifier": "UMLS:C0597277",
        "label": "potassium ion"
      },
      {
        "identifier": "RXCUI:1318915"
      }
    ],
    "type": [
      "biolink:SmallMolecule",
      "biolink:MolecularEntity",
      "biolink:ChemicalEntity",
      "biolink:PhysicalEssence",
      "biolink:ChemicalOrDrugOrTreatment",
      "biolink:ChemicalEntityOrGeneOrGeneProduct",
      "biolink:ChemicalEntityOrProteinOrPolypeptide",
      "biolink:NamedThing",
      "biolink:PhysicalEssenceOrOccurrent"
    ],
    "information_content": 53.4
  }
}

The empty node categories list in the knowledge graph returned by Retriever is presently causing problems for ARAX. Is it valid TRAPI, and an expected possible return value for a node in the knowledge_graph, to have a node's categories attribute be an empty list? We could kludge around this, but first I thought I would reach out, to better understand if this is normal or unexpected. Thank you!

[saramsey]

See ARAX issue 2682:
RTXteam/RTX#2682

[tokebe]

Ah, I see the cause here. Retriever only finds an edge based on implicit subclass reasoning, so it never gets back node information for the original query node.

@shuchenliu Any ideas? Can we query ES for just a node in the current data format?

[saramsey]

Hi Willow, it looks like this issue is fixed, is that right?

        "sn": {
          "ids": [
            "CHEBI:29103"
          ],
          "categories": [
            "biolink:ChemicalEntity"
          ],
          "set_interpretation": "BATCH",
          "constraints": [],
          "member_ids": [],
          "is_set": false
        }

[tokebe]

No, the underlying cause has not been fixed, @shuchenliu is working on a fix now.

Looks like you're looking at the query node when you mean to be looking at the knowledge node?

[saramsey]

Oops, sorry, thanks for your reply.

[saramsey]

I just saw the PR! @shuchenliu does this mean #143 is slated to have a fix deployed? Is it "live" on the Retriever dev endpoint? We are ready to test on our end, whenever we hear that it is appropriate to do so.

Many thanks for your work on this issue.

[shuchenliu]

@saramsey it's merged into main branch but that has not been built on the dev instance yet. I'll let you know when that's done - shouldn't take too long though

[shuchenliu]

@saramsey the changes are now live on the Retriever dev instance. One quick note is that the original query your provided won't fetch any results given the changes in the underlying ubergraph. If you substitute CHEBI:29103 with CHEBI:48927 however, there should be results returned and all nodes fetched back will be properly loaded.