Proposal: do not allow normalisation of concept level edge types?

[matentzn]

There are good arguments that edges at concept level should never be normalised, see everycure-org/matrix#1236 how biolink can separate these out.

For example, If we normalise diseases to Mondo, we inherit an insane amount of subclass of edges from dozens of disease terminologies, instead of simply using the one and only classification provided by Mondo which is internally consistent (does not have cycles etc). Furthermore other concept level relations like close match etc don't make sense to normalise at all.

Action items

• Decide if we should skip concept level edge types from normalisation to ensure those relations are consistent with their source ontologies

@kevinschaper @eKathleenCarter and @cbizon do you see where I am going with this? Does it make sense? Low priority and backlog?

[cbizon]

Yeah I agree - subclass of is a case where we should probably try to control things a bit. For instance, we could do something like 'outsource' the logical collection of these to ubergraph and then at integration time only accept subclass of if it comes from that source (or make sure that we pull that source directly rather than via one of the other graphs)

[matentzn]

Yeah I agree - subclass of is a case where we should probably try to control things a bit. For instance, we could do something like 'outsource' the logical collection of these to ubergraph and then at integration time only accept subclass of if it comes from that source (or make sure that we pull that source directly rather than via one of the other graphs)

Yes, good compromise.

[eKathleenCarter]

Yes, I think this sounds good and logical to me as well.