Ranking MVP2: the notion of path length matters

NCATSTranslator / Feedback

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

7 stars 0 forks source link

Ranking MVP2: the notion of path length matters #747

Closed TranslatorIssueCreator closed 3 months ago

TranslatorIssueCreator commented 7 months ago

Type: Suggestion

URL: https://ui.test.transltr.io/main/results?l=Potassium%20Ion&i=PUBCHEM.COMPOUND:813&t=3&r=0&q=66f83ee1-3054-4d7a-8e7b-35def1c1a267

ARS PK: 66f83ee1-3054-4d7a-8e7b-35def1c1a267

Steps to reproduce:

Screenshots:

sandrine-muller-research commented 7 months ago

In this query the top answer is an inferred query of "affected_by" that invlves a chain going through increasing the expression of 2 genes. Even if there is a lot of evidence of that chaining, is less relevant to me than the look up with ATP (path length = 1).

sierra-moxon commented 7 months ago

@webyrd - do you think you could help us sort out why the 2 hop has a better score than the 4 one-hops? CD70 in the ARAX interface seems like it as a direct relationship but is ranked 12 vs. VPS51 which has a 2 hop and is ranked 1.

maybe the direct edge doesn't score well, but the multi-hop path multiplier makes them rank higher? (e.g. 4 not great 2 hop paths == a better score)

sstemann commented 3 months ago

i ran this on Test (which is now Fugu), but you cannot use Potassium Ion, only Potassium. All results are one-hop.

https://ui.test.transltr.io/main/results?l=Potassium&i=CHEBI:26216&t=3&r=0&q=3cc31376-1333-46a0-9ab0-6499440ebf28

@sandrine-muller-research not sure if you have another case of this. i think there are a lot of questions regarding rank, score, evidence that are wrapped up with O&O. But without an actual example on this one there isn't much to pass off right now

sandrine-muller-research commented 3 months ago

Let's close this and I'll open another ticket if I find abother case (I have logged a few ranking issues already so perhaps we have already)