Open karafecho opened 2 years ago
Just some quick notes about how we can check whether this is likely to have much/any impact before we actually write any additional code.
I believe the issue to be addressed can be seen in the ARAX response for Query B.1. The observation is that many non-specific nodes (e.g., "cytokine", "MicroRNAs", "agonists") are appearing in the results. The hypothesis is that filtering on "novelty" node property in SemmedDB (eliminating nodes where novelty = 0
) would enrich for results that SMEs would care about.
The Service Provider has created a new API for semmeddb that includes the novelty score at https://biothings.ncats.io/semmeddb. So let's check a few of the B.1 results to see what their novelty scores are:
FMA:84050
, which synonym matcher says is UMLS:C0079189
, which can be searched using https://biothings.ncats.io/semmeddb/query?q=subject.umls:C0079189%20OR%20object.umls:C0079189, which says novelty: 1
. So novelty would not be helpful in eliminating "cytokine"MESH:D035683
-> UMLS:C1101610
-> https://biothings.ncats.io/semmeddb/query?q=subject.umls:C1101610%20OR%20object.umls:C1101610 -> novelty: 1
agonists: MESH:Q000819
-> UMLS:C0243192
-> https://biothings.ncats.io/semmeddb/query?q=subject.umls:C0243192%20OR%20object.umls:C0243192 -> novelty:0
I hope I've illustrated how the semmeddb novelty score currently can be accessed. Perhaps a SME or analyst with workflow B could do more spot checking on the nodes they would consider to be "too generic"?
This issue is to follow up on Andrew Su's suggestion to consider ways to leverage SemMedDB's "Novelty" constraint by setting it to Novelty=0 either in the query itself or via another approach.