KG2.8.0 edges with a qualified_object_direction, but no qualified_predicate

[amykglen]

I noticed that this edge in KG2.8.0c (which comes from the KG2.8.0pre edge with ID UMLS:C0043552---SEMMEDDB:stimulates---None---None---increased---UMLS:C0006809---SEMMEDDB:) has a qualified_object_direction, but no qualified_predicate:

{
   "predicate":"biolink:regulates",
   "knowledge_source":[
      "infores:semmeddb"
   ],
   "publications_info":"{'PMID:33316364': {'publication date': '2020 Dec 11', 'sentence': 'Both, EOOK and camphor inhibited all articular parameters induced by zymosan.', 'subject score': 1000, 'object score': 1000}}",
   "kg2_ids":[
      "UMLS:C0043552---SEMMEDDB:stimulates---None---None---increased---UMLS:C0006809---SEMMEDDB:"
   ],
   "subject":"MESH:D015054",
   "qualified_object_direction":"increased",
   "id":"27104526",
   "object":"CHEMBL.COMPOUND:CHEMBL1097205",
   "publications":[
      "PMID:33316364"
   ]
}

I don't think this is valid, at least according to the predicate transformations Sierra provided? those transformations seem to suggest this edge should have a qualified_predicate of "causes" and an object_aspect of "activity_or_abundance".

it looks like there are 2.4 million such edges in KG2.8.0c (that have a qualified_object_direction but do not have a qualified_predicate) based on this neo4j query:

match (n)-[e]->(m) where e.qualified_predicate is null and e.qualified_object_direction is not null return count(distinct e)

based on the KG2pre ID of the specific edge reported above, it seems that the qualified_predicate was already missing in KG2pre vs. being lost during the KG2c build process, so that's why I'm writing up this issue in this repo.

[amykglen]

it looks like all of the 2.4 million affected edges have the predicate 'regulates':

match (n)-[e]->(m) where e.qualified_predicate is null and e.qualified_object_direction is not null return distinct e.predicate, count(distinct e)

I wonder if it would be worth adding a temporary patch in the KG2c build code that fixes this (by simply giving all such edges a qualified_predicate of "causes" and an object_aspect of "activity_or_abundance")... otherwise there appear to be only 15k 'qualified' edges in KG2.8.0 that could be returned for queries that use qualifiers.

perhaps @sundareswarpullela would be interested in writing such a patch?

[sundareswarpullela]

Sure! I can pick this up @amykglen

[acevedol]

Thank you for the alert, Amy! I'll make the correction in KG2pre

[amykglen]

great, thanks @sundareswarpullela! I created an issue in the RTX repo for the patch so you can link your commits to it: RTXteam/RTX#1942

[acevedol]

A lot of entries in predicate-remap.yaml were missing qualified_predicate, so I added those. Hopefully, this solves this issue. I will verify in next build.