filtered out edges with same subject and object from results

NCATSTranslator / Feedback

A repo for tracking gaps in Translator data and finding ways to fill them.

7 stars 0 forks source link

filtered out edges with same subject and object from results #867

Open cbizon opened 3 months ago

cbizon commented 3 months ago

Query: what decreases ACE? Environment: TEST (EEL) PK: 2ea30f9b-871e-43e5-8f3b-ea4b43d8ee17

This has "decreases activity or abundance of" implied by "increases activity of"

Seems to be from Unsecret.

sierra-moxon commented 3 months ago

@kaiwenho - do you think my label of "Guppy" work is appropriate here, or is this for the next phase? :)

kaiwenho commented 3 months ago

@sierra-moxon - the label looks good to me!

I notice that for the inferred answer (Captopril from Unsecret) , one of the component edges (ACE causes decreased activity ACE) is missing in the UI: from UI:

In ARAX, the edge (ACE causes decreased activity ACE) is displayed:

Hence, the complete chain for this answer from Unsecret is Captopril - causes increased activity -> ACE - causes decreased activity-> ACE. But somehow, UI does not show the second component edge. @gprice1129 - could you please take a look at this answer? Thanks!

gprice1129 commented 2 months ago

Right now the UI doesn't support self-referencing edges... Can somebody with more biological background speak to if this makes sense? Essentially this would mean that as long as anything increases or decreases the activity of ACE... Then the activity of ACE will be decreased.

Genomewide commented 2 months ago

I think this is something that Unsecret needs to look at. I think it slipped through their filter. Also, I can't get this same issue to show up though with the KP 2ea30f9b-871e-43e5-8f3b-ea4b43d8ee17

I also think that we should not show in this case.

@kaiwenho @webyrd This looks odd. It smells like a logical step that breaks when there is self-reference.

kaiwenho commented 2 months ago

The self-referencing edges are filtered out from mediKanren server. The current view in CI -

sstemann commented 2 months ago

i feel a little confused, in Test/Fugu 763bf762-1d10-4295-9948-5ed260111fb2 - Unsecret does not have self referencing edges for this example. Maybe there are other examples with the self referencing.

Can someone explain what is being changed for Guppy?

kaiwenho commented 2 months ago

Can someone explain what is being changed for Guppy?

@sierra-moxon The issue was tested on TEST (EEL) which should be Prod (ELL) now. I guess because Unsecret updated the ranking algorithm on TEST (Fugu), the path containing self-referencing edge was excluded with a relatively low score and did not get returned. However, there might be other self-referencing edges as a look-up or component edges get returned in other queries, so implementing a filter, which is the change in CI (Guppy), is safe.

sstemann commented 1 month ago

looks to be still resolved in Test https://ui.test.transltr.io/results?l=ACE%20(Human)&i=NCBIGene:1636&t=2&r=0&q=d1ae79ae-c654-4cac-9df3-67e4d7f926ba