search

NCATSTranslator / Feedback

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

SemMedDB incorrectly returns Methimazole as a treatment for hypothyroidism #806

Open khanspers opened 4 months ago

khanspers commented 4 months ago

Query: What drugs may treat conditions related to:Hypothyroidism? https://ui.test.transltr.io/main/results?l=Hypothyroidism&i=MONDO:0005420&t=0&r=0&q=6810d92b-611e-488e-b2cc-883cdd7f772e

Methimazole is returned as a treatment for hypothyroidism, with text-mined results from SemMedDB. Methimazole is used to threat hyperthyroidism, and can therefor cause hypothyroidism (see https://www.ncbi.nlm.nih.gov/books/NBK545223/).

Screen Shot 2024-06-25 at 2 53 16 PM

andrewsu commented 4 months ago

With SemMedDB-related issues, my first question is whether this is related to Service Provider's ingestion of SemMedDB or RTX-KG2's? To figure this out, I checked out the ARAX UI at https://arax.ci.transltr.io/?r=6810d92b-611e-488e-b2cc-883cdd7f772e

There, I see that this answer is coming from BTE:

image

To see the specific record(s) from the SemMedDB API that are being cited, I get the UMLS IDs by clicking the nodes in the ARAX UI (hypothyroidism: C0020676, methimazole: C0025644). Then I query the SemMedDB API using https://biothings.transltr.io/semmeddb/query?q=subject.umls:C0025644 AND object.umls:C0020676 (for completeness, I also checked https://biothings.transltr.io/semmeddb/query?q=object.umls:C0025644 AND subject.umls:C0020676)

I see the methimazole - treats - hypothyroidism edge has a pmid_count of 32, while methimazole - causes - hypothyroidism has a pmid_count of 118. Based on these results, the most reasonable way of filtering semmeddb to remove this edge would be the approach to look at conflicting results via opposing predicates (described in https://github.com/NCATSTranslator/Feedback/issues/392).