MVP1 Inferred Edges Return with Semmed edge as a support graph

[sstemann]

Test UI > MVP1 > Cystic FIbrosis

https://ui.test.transltr.io/main/results?l=Cystic%20Fibrosis&i=MONDO:0009061&t=0&r=0&q=01f390c0-0d65-4dcc-a6ea-4b1749ec10b2

Find the result Iron

ARAX - Treats - no support graph

BTE - Treats_or_applied_or_studied_to_treat - no support graph

Unsecret - Treats - no support graph

ARAGORN had 0 results, Iron is not in the Improve Set

[cbizon]

The predicate for the semmed edge itself should be "treats or applied or studied to treat". I think that this has not been done in RTX KG2, and that is causing the error in both ARAX and UnSecret.

For BTE the problem is different - rather than returning the (correct) semmed edge as a direct lookup bound to the query graph edge, it should be returning an inferred treats edge with the semmed edge as a support graph.

[saramsey]

This should be fixed in Fugu,

https://github.com/RTXteam/RTX-KG2/blob/9075635c3beed053d226c69db0a6f021064dbf9f/maps/predicate-remap.yaml#L4490-L4495

tagging @ecwood who can confirm that this is in the KG2.10.0 build that will be (hopefully) deployed to ITRB TEST by 7/26 for the Fugu release.

[saramsey]

For this to be already fixed in ITRB TEST at the present time, it would have to have been in the Eel release; but there was (AFAIK) no new release of RTX-KG2 in the Eel release, so we have to wait for Fugu, I think.

[sstemann]

there is an Eel release of KG2 in Test versioned v2024.06025.1-test. But since this is occurring in Test, looks like it still needs to go to Eel

[saramsey]

Hi @sstemann yes, this fix made it into Fugu, but not Eel. Here you can see the fix in Fugu:

[saramsey]

there is an Eel release of KG2 in Test versioned v2024.06025.1-test. But since this is occurring in Test, looks like it still needs to go to Eel

Apologies for my lack of precision. What I meant to convey is that in the Eel release of Translator, due to staffing turnover (loss of a team member who was the point person at the time on RTX-KG2 builds), Expander Agent wase not able to put out a new version of RTX-KG2. Thus, the "treats" fix (for the RTX-KG2 KP) is in Fugu, not in Eel.

[sstemann]

these two edges for the same direct path have the same 11 pubs from Semmed

is this by design?

Free text search > Iron

https://ui.test.transltr.io/main/results?l=Cystic%20Fibrosis&i=MONDO:0009061&t=0&r=0&q=9c685631-2386-4ade-a97a-8f939e6b2f54

ARAX says Treats with the 11 pubs

BTE says treats_or_applied_or_studied_to_treat with the same 11 pubs

[sstemann]

i'm not sure who said Guppy and who said Fugu @saramsey @colleenXu

[colleenXu]

(1) It appears that both edges have the same primary knowledge source (semmeddb), which would explain the same data. Different formatting/predicates may be due to the aggregator source (rtx-kg2 vs biothings/service provider/bte).

(2) It looks like Sierra assigned those labels. I don't recall saying anything about what release our team's fix would be in (Jackson brought this up today in our bte-itrb Slack channel).

[sierra-moxon]

this comment is how the labels were assigned: "there is an Eel release of KG2 in Test versioned v2024.06025.1-test. But since this is occurring in Test, looks like it still needs to go to Eel"

[sstemann]

@colleenXu thank you for the review and update, any plan for BTE to change the format of the predicates?

[colleenXu]

@sstemann

There is now a fix deployed to BTE CI. It should now return "an inferred treats edge with the semmed edge as a support graph" - which is what Chris Bizon recommended.

[sstemann]

looks resolved in Prod