BTE creating self-edges via subclassing

[tokebe]

Example: NCATSTranslator/Feedback#907

Something is wrong in node-expansion and subclass handling, likely a missing check somewhere that should be causing edges to be dropped.

[tokebe]

Looks like this is being caused by the following process:

• Node is expanded from originalID -> descendantID
• BTE gets record(s) back that shows descendantID -> has_phenotype -> originalID
  • This misses edge-management self-edge avoidance because originalID != descendantID
  • This also misses self-subclass avoidance because it's not technically a self-subclass
• BTE rebinds descendantID -> has_phenotype -> originalID to originalID -> has_phenotype -> originalID with underlying support graph(s) originalID -> has_subclass -> descendantID -> has_phenotype -> originalID

This should be possible to catch at support graph creation time -- I'm double-checking for potential knock on effects of catching it there.

[rjawesome]

Note: there also needs to be tracking of which QNodes subclassing should be applied to, otherwise subclassing created for one QNode (ie. n0) could be applied to a more specific node bindings for a more specific QNode (ie. n1)
Example Query

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids":["MONDO:0006497"]
                },
                "n1": {
                    "categories": ["biolink:DiseaseOrPhenotypicFeature"],
                    "ids": ["MONDO:0016215"]
               }
            },
            "edges": {
                "e0": {
                    "subject": "n0",
                    "object": "n1",
                    "predicates": ["biolink:has_phenotype"]
                }
            }
        }
    }
}

Result (subclassing changes n1 from MONDO:0016215 to MONDO:0006497 even though n1 is explicitly set to MONDO:0016215)

"results": [
      {
        "node_bindings": {
          "n0": [
            {
              "id": "MONDO:0016215",
              "attributes": []
            }
          ],
          "n1": [
            {
              "id": "MONDO:0006497",
              "attributes": []
            }
          ]
        },
        "analyses": [
          {
            "resource_id": "infores:biothings-explorer",
            "edge_bindings": {
              "e0": [
                {
                  "id": "MONDO:0016215-has_phenotype-MONDO:0006497-via_subclass",
                  "attributes": []
                }
              ]
            },
            "score": 0.7308028298005093
          }
        ]
      }
    ]
  },

[colleenXu]

HOLD ON

@rjawesome @tokebe: See my recent post where I wonder if the "duplicate second path" in the Feedback issue is actually coming from Aragorn NCATSTranslator/Feedback#907 (comment)

If that is the case, then the self-edge doesn't seem to be showing in the UI. Which would mean this issue of self-edges is less urgent/breaking.

---

There is a "true duplicate paths in the UI" situation going on, but that's happening because of how the ARS/UI handles our edges with subclassing support-graphs vs ones without NCATSTranslator/Feedback#776 (comment). This does make me wonder if Translator will ask us to undo our subclassing support-graph wrapping...

---

FYI Another example with a self-edge (CI, ran earlier today): https://arax.ci.transltr.io/?r=3c00776f-a6aa-473e-907f-3dec1367a566

Screenshot from ARAX-CI view

But I didn't notice any duplicate paths issues in the UI view. There were similar paths (highlighted in the screenshot below), but I think they were from Aragorn, based on my analysis of the Test-instance version here.

This is why I started to suspect that the self-edge wasn't the source of the duplicate paths in 907 either...

Screenshot of UI-CI view

[tokebe]

I'm pretty sure regardless of ARS/UI-level merging errors, BTE is still creating self-edges via subclassing that we don't want? @colleenXu unless there's a reason to keep these self-edges, @rjawesome don't pause any active work on this issue.

It's fine if there's simultaneous effort outside of this issue toward problems that are related to NCATSTranslator/Feedback#907 but aren't related to this issue -- we can separately track a BTE-level problem here and assignees to the Feedback issue can decide whether our efforts are related/causal to 907.

[colleenXu]

I tested the two linked queries and things looked good: the paths containing self-edges have been removed. I did see a different in scores/rank, often a slight decrease for the results that previously had self-edges.

But in the maturity-onset diabetes case below, the rank drastically dropped, but the score only decreased a little.

cerebral palsy (MONDO:0006497)

Before (ARAX-CI viewing, used BTE-test)

After: cerebral-palsy-issue-850-testing.json.zip

maturity-onset diabetes of the young (MONDO:0018911)

Before (ARAX-CI viewing)

After: MoDY-issue-850.json.zip

---

I also did a quick test of a MVP2 query and didn't notice anything off about its response (increases ADRB2 - NCBIGene:154).

[colleenXu]

This issue is also tracking a fix to "edge-case query failures due to an error in subclassing".

Info from #789 (comment)

Context: the error revealed a bug in node-expansion, coming from MONDO ontology having non-MONDO ID info - including mismatched ID namespaces for parent-child pairs. HP has this issue too.

The first test query used an UBERON ID, which had some child info in those files with a different ID namespace.

The solution is to remove those mismatched pairs.

Jackson has put a fix into the subclassing-fix branches (deployment tracked in #850)

• fix subclassing, self graphs bte_trapi_query_graph_handler#208
• https://github.com/biothings/node-expansion/tree/subclassing-fix

[tokebe]

Relevant changes deployed to Prod.