search

NCATSTranslator / Feedback

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

All contain xxx--treats -- xxx --Phenotype of-- with no Genes as intermediary nodes #905

Closed sstemann closed 3 months ago

sstemann commented 3 months ago

Retested today on PROD: I could not find the same problematic paths under Aripiprazole but we do have problematic paths in my opinion (not very clever reasoning, often leading to bad results IMO):

image

They all contain xxx--treats -- xxx --Phenotype of-- with no Genes as intermediary nodes

Originally posted by @sandrine-muller-research in https://github.com/NCATSTranslator/Feedback/issues/357#issuecomment-1751203852

sstemann commented 3 months ago

these seem to be BTE results. I believe this is by design but will let BTE respond.

https://ui.transltr.io/main/results?l=Autism%20Spectrum%20Disorder&i=MONDO:0005258&t=0&r=0&q=a85f95bf-80b3-49e6-b42a-cba033fec6f6

andrewsu commented 3 months ago

I'll take this one too (just to keep @tokebe on some other priorities)...

tokebe commented 3 months ago

Probably best to direct these kinds of issues to @andrewsu in the future; if it's clearly evident of a bug by all means direct it to me, but otherwise I think it might make more sense for Andrew to receive and delegate these as appropriate.

andrewsu commented 3 months ago

Yes, everything is operating by design. In https://github.com/biothings/bte_trapi_query_graph_handler/blob/main/data/templateGroups.json#L1-L16, we define the specific templates that the creative mode query is expanded to. The TRAPI for each template is in https://github.com/biothings/bte_trapi_query_graph_handler/tree/main/data/templates/Drug-treats-Disease. We of course know that the selection of templates and our scoring function can be improved. I'm going to close this issue, but feel free to repoen if anyone has a specific suggestion. (I'll also note that BTE reports sirolimus with rank 117...)