gene -- causes -> phenotype edge from uniprot?

[amykglen]

Sierra was a bit concerned at today's data modeling call to see that the KG2 API returns a uniprot 'causes' edge for the query NCBIGene:351 (APP) -- related_to --> HP:0100543 (Cognitive impairment)

I traced this to the edge with this ID in KG2.10.0pre:

UniProtKB:P05067---biolink:causes---None---None---None---OMIM:104300---identifiers_org_registry:uniprot

(which was returned from the KG2 CI API for this query due to concept subclass reasoning)

does anything seem fishy here? or do we expect to have this sort of edge from uniprot?

[edeutsch]

In principle UniProtKB does contain such information:
https://www.uniprot.org/uniprotkb/P05067/entry#disease_variants

[saramsey]

See this code:

RTX-KG2/convert/uniprotkb_dat_to_kg_jsonl.py

Lines 202 to 212 in dbeeb16

	if 'disease' in record_dict:
	for disease_rec in record_dict['disease']:
	mims = REGEX_MIM.findall(disease_rec)
	for m in mims:
	mp = REGEX_PUBLICATIONS.findall(disease_rec)
	pubs = [fix_publications(pub) for pub in mp]
	e = kg2_util.make_edge_biolink(curie_id,
	kg2_util.CURIE_PREFIX_OMIM + ':' + m,
	kg2_util.EDGE_LABEL_BIOLINK_CAUSES,
	UNIPROTKB_PROVIDED_BY_CURIE_ID,
	update_date)

[saramsey]

OK to close this issue, @amykglen ?

[amykglen]

yes, thank you - I suppose we should let Sierra know:

@sierra-moxon - just letting you know that we investigated this UniProt 'causes' edge in KG2 that you brought up on the Data Modeling call in October, and everything looks ok to us. basically the raw form of the edge being returned was UniProtKB:P05067--causes-->OMIM:104300 (Alzheimer's disease), which is actual info that UniProt contains (see above comments). and this edge was returned for the original query of NCBIGene:351 (APP) -- related_to --> HP:0100543 (Cognitive impairment) since UniProtKB:P05067 is a synonym for NCBIGene:351, and Alzheimer's disease is a descendant concept of Cognitive impairment (at least according to our concept subclass reasoning, which is based on MONDO and HP subclass_of edges).

also interestingly, in CI, this causes edge is no longer returned - the only 'causes' edges now linking NCBIGene:351 and HP:0100543 are from SemmedDB - perhaps due to changes in the underlying KG2 graph, node normalization, or other parts of the system..