semmed sentence edge attributes

[rjawesome]

biothings/biothings_explorer#833

Needs semmedb's response mapping to ingest the edge attributes.
aka, add the following to all response mappings

edge-attributes: edge-attributes

Can also support saying subject/object location, if subject/object text is retrieved from the semmed api as a field

[colleenXu]

@rjawesome

Is this ready for testing? The linked issue isn't in the "Awaiting Review" column...

Also, on "subject/object location can be supported"....

• looks like lines 20-27 need to be uncommented in order to test this, right?
• it looks like the values produced would be in the text-mining format, right? (looks like: pipe-limited string, 2 int values for start and end character index)

[rjawesome]

Is this ready for testing? The linked issue isn't in the "Awaiting Review" column...

I wanted to do a basic test (which would require me to modify the smartapi yaml to ingest the edge attributes). Hopefully I can complete that today and then mark it for review.

looks like lines 20-27 need to be uncommented in order to test this, right?

Yes, but this won't work until the smartapi yaml is updated to request the subject_text and object_text fields from the SemmedDB BioThings API (otherwise it throws an error since those fields are missing).

it looks like the values produced would be in the text-mining format, right? (looks like: pipe-limited string, 2 int values for start and end character index)

Correct.

[rjawesome]

@colleenXu
More Notes (I am now marking this as ready to review):
When adding the edge-attributes to the smartapi yaml (response mapping), it needs to be escaped with backticks and quotes like so (due to how jsonata works):
edge-attributes: "`edge-attributes`"

In addition to requesting predication.subject_text and predication.object_text for getting the end/start characters, we also need to request predication.predication_id for the attribute value. (this is in the parameters.fields of every operation in the smartapi yaml)

[colleenXu]

@rjawesome

(1) I originally proposed using semmeddb_publication_info: predication for the response-mapping (now I wonder if ref_text_mining is better, to match previous work).

Is there a reason you went with edge-attributes: "`edge-attributes`" instead (both different keyword and value)?

• Specifically, we've been using edge-attributes for Multiomics/Text-Mining to ingest the json containing their TRAPI edge-attributes without mutating them (previous issue)...so from my POV, it seems confusing to use the same keyword here.
• I was imagining a setup closer to the previous work where the original field (predication) could still be used as a value.

(2) In my original proposal and this branch, I changed parameter.fields to retrieve the entire predication json. Does this work? I see your comments that fields are missing (subject_text, object_text, predication_id), which I'm confused by...

[rjawesome]

@colleenXu

(1) We can keep semmeddb_publication_info: predication as well if any predication info was to be used directly. The JQ transformer that I wrote transforms the publication info directly into the TRAPI attribute format (it keeps the predication field as is while creating the edge-attributes field to store the TRAPI-formatted attributes). Hence, it would be ingested as edge-attributes so BTE knows it is already in TRAPI format.

ie. edge-attributes would contain something like this

{
              "attribute_type_id": "biolink:has_supporting_study_result",
              "attributes": [
                {
                  "attribute_type_id": "biolink:supporting_text",
                  "value": "CONCLUSIONS: Induction of the ARE-Nrf2 pathway by zinc provides powerful and prolonged antioxidation and detoxification that may explain the beneficial effects of zinc observed in the treatment of age-related macular degeneration (AMD)."
                },
                {
                  "attribute_type_id": "biolink:publications",
                  "value": "PMID:16723490"
                },
                {
                  "attribute_type_id": "biolink:subject_location_in_text",
                  "value": "50|54"
                },
                {
                  "attribute_type_id": "biolink:object_location_in_text",
                  "value": "34|38"
                }
              ],
              "value": "166335624"
            },
            {
              "attribute_source": "infores:text-mining-provider-targeted",
              "attribute_type_id": "biolink:has_supporting_study_result",
              "attributes": [
                {
                  "attribute_type_id": "biolink:supporting_text",
                  "value": "There was gender difference for the protective effect of zinc against diabetes-induced pathogenic changes and the up-regulated levels of Nrf2 and MT in the aorta."
                },
                {
                  "attribute_type_id": "biolink:publications",
                  "value": "PMID:23536959"
                },
                {
                  "attribute_type_id": "biolink:subject_location_in_text",
                  "value": "57|61"
                },
                {
                  "attribute_type_id": "biolink:object_location_in_text",
                  "value": "137|141"
                }
              ],
              "value": "168073659"
            },

(2) I missed the branch on the translator registry repo. That covers all the predication fields, however, the changes to get the edge-attributes are still needed.

[colleenXu]

@tokebe @rjawesome

Sorry for the delay in testing.

It looks good except for 1 thing: it's not respecting the edge max limit of 50 unique sentences/publications, aka only make 50 of these special edge-attributes. Could there be a problem when multiple records get merged into 1 edge?

You can use the translator-api-registry branch I've made. The response-mapping should match Rohan's specs now.

Example

add to override: "1d288b3a3caf75d541ffaae3aab386c8": "https://raw.githubusercontent.com/NCATS-Tangerine/translator-api-registry/refs/heads/semmeddb_publication_refactor/semmeddb/smartapi.yaml",

Query:

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids": ["NCBIGene:6752", "NCBIGene:2641"],
                    "categories": ["biolink:Gene"]
                },
                "n1": {
                    "categories": ["biolink:Polypeptide"]
                }
            },
            "edges": {
                "e01": {
                    "subject": "n0",
                    "object": "n1",
                    "predicates": ["biolink:affected_by"]
                }
            }
        }
    }
}

Look at edge b6d8cea8e750400d3224490240a63871. It has 90 edge-attributes: 3 for evidence_count/KL/AT and 87 for the 87 publications. It should have been cut off at 50.

Other examples are:

• 03bb8b167fae7431a88b50a593b029b0 (92 publications, 95 edge-attributes)
• 3dbd5535b00e8123ecef8f89ca29a685 (57 publications)
• 023c3ff7d5d3d9fa74d46a8a4ea675bd (73 publications)

[rjawesome]

I'll try to take a look at it this week.

[rjawesome]

@colleenXu The amount of publications in the Semmed edge attributes should be capped at 50 now (the max edge attributes in the example decreases to 53)

[colleenXu]

@tokebe

I've tested using the latest main branches + this PR + biothings/bte-server#45.
It looks good! The capping seems to be working now.

I think this is ready to put onto CI for testing this weekend...