How to handle NOT edges when subclass_of reasoning is in play

[amykglen]

so our new subclass_of reasoning interacts in interesting ways with the handling of exclude=True (NOT) edges... here's the situation:

consider this QG:

there are 16 chemical substances in KG2c that both treat and predispose MONDO:0004985 (bipolar disorder):

['(+/-)NICOTINE', 'ALCOHOL', 'ANAKINRA', 'ATOMOXETINE HYDROCHLORIDE', 'Antidepressants', 'CHOLECALCIFEROL', 'FOLIC ACID', 'LEVOTHYROXINE SODIUM', 'LITHIUM ION', 'METHYLPHENIDATE', 'NITRIC OXIDE', 'OMEGA-3 FATTY ACIDS', 'TRASTUZUMAB EMTANSINE', 'agonists', 'antioxidant', 'antipsychotic agent']

(as a side note: the quality of this list seems a little questionable? but that's separate from this issue.. 🙂)

before subclass_of reasoning was implemented, all 16 of those chemical substances were excluded from the results for this query, due to the NOT edge.

but now that we have subclass_of reasoning, two of those chemical substances (Antidepressants and LITHIUM ION) are still actually appearing in the results for this query, because they have a treats edge to a child of bipolar disorder (bipolar II disorder - MONDO:0000693), but no predisposes edge to that child. https://arax.ncats.io/?r=16819

is this the way we want it to work? or should those treats edges between Antidepressants/LITHIUM ION and bipolar II disorder also be blown away, since Antidepressants and LITHIUM ION predispose bipolar disorder (the parent)?

I suppose I'm thinking our current handling is best, but thought I'd raise the issue for anyone else's input..

[edeutsch]

Tough question. I think I agree current handling is best, but I'm not at all certain about that.

[finnagin]

@edeutsch @amykglen is this okay to close?

[edeutsch]

This is quite stale and I think this is okay to close.