What Chemical Substances Cause AKU MONDO:0008753 - two Actors Return Results

[sstemann]

Query: AKU PK: f7600636-6249-4abb-a592-cf7f89de45aa GO: MONDO:0008753 Results Tracking Sheet Control: Homogentisic Acid

Responses From: ARAX (Expander Agent) BTE (Exploring Agent)

[marcdubybroad]

The Genetics KP doesn't contain 'biolink:ChemicalSubstance' entities, so the 0 results is expected.

[rtroper]

Currently, our EHR risk KP relates drug (rather than chemical substance) to disease. Also, we don't have alkaptonuria, but we might be able to add it in the update we're working on (along with many other diseases we're adding).

[bill-baumgartner]

Text Mining Provider does not currently target this specific relation. I've added a ticket to our backlog to consider this as a future target.

[cbizon]

Ranking-agent is now returning results. We were not previously querying BTE for object based queries.

[vdancik]

We are not able to handle "causes" predicate at the moment

[pg427]

Explanatory Agent does not support this specific relation, currently.

[brettasmi]

Hi Everyone,

I am posting the materials that Sergio and I described earlier on the call.

• aku_which_compound.json.txt contains the TRAPI query to find homogentisic acid via the metabolic pathways, essentially asking, "Which compounds are involved in reactions have enzymatic activity involving proteins that are products of genes associated with AKU?"
• aku_treatments.json.txt contains a TRAPI query to walk back out of the metabolic pathways with the new knowledge of homogentisic acid. The query asks, "Show me compounds that affect genes that have protein products that have enzymatic activity in reactions that involve homogentisic acid

(sorry about the .txt naming; GitHub does not allow the .json file extension for attachments)

Here's an illustration of the network from the SPOKE neighborhood explorer. The blue path denotes the first query and the orange path denotes the second query.

Some thoughts:

• The two neighboring biolink:MolecularAcitivity nodes are a bit odd, but the KEGG.REACTION and KEGG.EC are both molecular activity in biolink.
• I think pathways are going to be hard to generalize in Translator because I imagine they will be subject to strong opinions in modeling, and these queries are thus probably specific to the schema of SPOKE. You can find an image of the SPOKE schema here if it helps you wrap your head around what's happening in the queries.

aku_treatments.json.txt aku_which_compound.json.txt

[sierra-moxon]

Discussion is also ongoing in the biolink-model repo: https://github.com/biolink/biolink-model/issues/478 (note comments about MolecularActivity). We plan to do a bit of a write up and be sure to discuss further with interested folks. :)

[cbizon]

@brettasmi that is very cool. Is the thought that improving agent could get the "causes" query, and figure out that this kind of longer query could be useful and run it? Or would you expect that the ARS would send you this longer query in the first place?

[colleenXu]

I ran an updated query below through the ARS, PK: f4d79bc5-390f-43cd-8342-ada49ed227f1

However, BTE does not find a ChemicalEntity that causes the Disease (instead it finds a ChemicalEntity that treats the Disease).

The updated query could be something like this (it's missing the workflow section but the rest is TRAPI v1.2):

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "ids": ["MONDO:0008753"],
                    "categories":["biolink:Disease"]
                },
                "n1": {
                    "categories": ["biolink:ChemicalEntity"]
                }
            },
            "edges": {
                "e0": {
                    "subject": "n0",
                    "object": "n1"
                }
            }
        }
    }
}