search

RobokopU24 / Feedback

Feedback on the ROBOKOP project
https://robokop.renci.org
0 stars 0 forks source link

Lots of bidirectional has_phenotypes #160

Closed cbizon closed 1 year ago

cbizon commented 1 year ago

In robokop: 

match (a:`biolink:NamedThing`)-[x:`biolink:has_phenotype`]->(b)-[y:`biolink:has_phenotype`]->(a) return * limit 10

Returns plenty: image

All of them look like one direction is coming from the SRI graph and the other is coming from hetio. That suggests to me that one of them is being inverted on load (probably hetio?)

EvanDietzMorris commented 1 year ago

Unfortunately it's not as simple as an inversion. Hetio has 16,570 has_phenotype edges, 7,459 of them are coming from MEDLINE Co-Occurrences (MRCOC), which if I understand correctly are literature co-occurrences that don't really say anything about directionality, but Hetio has necessarily assigned one (probably arbitrarily). All of the other examples of has_phenotype in Hetio I've looked at seem to have the correct directionality.

Should we throw out the MEDLINE edges? Change the predicate to correlated_with for those?

cbizon commented 1 year ago

I think throw them out

EvanDietzMorris commented 1 year ago

There is a filter in the parser now that removes them.