search

NCATSTranslator / Feedback

A repo for tracking gaps in Translator data and finding ways to fill them.
7 stars 0 forks source link

ChEBI Name insulin (human) for ChEBI ID CHEBI:5931 is labelled "Insulin Beef" by the ARAs #973

Open sstemann opened 1 week ago

sstemann commented 1 week ago

Search "What genes may be upregulated by Insulin (human)?" CHEBI 5931 and in Test, the results are for "insulin beef"

https://ui.test.transltr.io/results?l=Insulin%20Human&i=CHEBI:5931&t=3&r=0&q=466f8a3b-43c1-40e3-a257-0280e357f4ae

image

It looks to affect all results that make it to the UI. there are 878 results total, and when you text search for "insulin beef" all 878 are returned.

In Prod, the search is for "Insulin" (without the human part), but it's the same CHEBI 5931.

The results however, aren't for "insulin beef" which seems like the behavior we do want. If you search "beef" no results match.

https://ui.transltr.io/results?l=Insulin&i=CHEBI:5931&t=3&r=0&q=b0d3c65f-5887-4a91-8188-c70853834b5d

image

CHEBI 5931 also shows up in MVP1 Results as Insulin Beef in Test (but not in Prod).

https://ui.test.transltr.io/results?l=Diabetes%20Mellitus&i=MONDO:0005015&t=0&r=0&q=d7e7118c-4758-42de-933c-804b3fb6aa62

image

UI team has reviewed and confirmed that the UI uses node names as provided by ARAs. So I'm wondering why ARA's are using "insulin beef" when the ChEBI page is "Insulin (human)". If you investigate the PKs for the test runs, it's happening for ARAGORN, ARAX, BTE and Unsecret.

andrewsu commented 1 week ago

How odd... My first guess was that this was an issue with node norm, but all instances appear to be reporting "insulin" and not "insulin beef"

I'm a bit stumped...

EDIT: from @gaurav via slack:

The bug only shows up if you have drug_chemical_conflate turned on: https://nodenorm.transltr.io/get_normalized_nodes?curie=CHEBI:5931&drug_chemical_conflate=true --> Insulin https://nodenorm.test.transltr.io/get_normalized_nodes?curie=CHEBI:5931&drug_chemical_conflate=true --> Insulin beef https://nodenorm.ci.transltr.io/get_normalized_nodes?curie=CHEBI:5931&drug_chemical_conflate=true --> Insulin beef I think the bug is older than Hammerhead, but I think there was another bug that hid it until Hammerhead came out

MarkDWilliams commented 1 week ago

It should be the case that the ARS replaces any names that don't match the preferred ones returned by node norm, except in cases where an ARA has a name for a node and node norm does not. I'll take a closer look. Not sure yet what's happening.