search

NCATSTranslator / TranslatorArchitecture

MIT License
9 stars 11 forks source link

edge merge #73

Closed cbizon closed 2 years ago

edeutsch commented 2 years ago

So one complication with this:

When an ARA obtains multiple edges with the same subject, predicate, qualifiers, object, and original source from KPs, it should represent these as a single edge in the knowledge_graph component of a ReasonerAPI message.

In the RTX-KG2 knowledge graph, some identical edges are already merged in the knowledge graph and these are reported with multiple sources. For example, see the related_to edge in this result: https://arax.ncats.io/?r=41651 image image

What is the merging policy when merging this edge with a edge from a different KP that only has one of the 3 knowledge sources, say. Also, here we use three separate attributes, but it really could have been a single attribute with a list of 3 original sources. Thoughts on that? Should we clarify in the language to guide implementation?

cbizon commented 2 years ago

Modified based on @edeutsch's comment above, and the subsequent Architecture Call discussion.

@kennethmorton @ehinderer @edeutsch @vdancik @CaseyTa @suihuang-ISB you may wish to review to make sure that your approval still stands.

edeutsch commented 2 years ago

okay, thanks. Also tagging @saramsey since this impacts RTX-KG2.

edeutsch commented 2 years ago

At the risk of seeming overly pedantic, should we replace "original source" with "primary source" in the PR?

cbizon commented 2 years ago

@edeutsch I made a little change "original"->"original/primary". Each edge should have one or the other only - My understanding is that the EPC group is probably going to remove one or the other (and then every edge should have that one), but I don't know which one is going to stick around.

kennethmorton commented 2 years ago

I still approve of the stated plan. I believe this has been discussed in various working groups, but documenting here for historical sake, I believe RTX-KG2 is currently merging in a manner that is inconsistent with the stated plan.

cbizon commented 2 years ago

This one is ready to merge, but I think I'll hold it open until EPC provides feedback. @mbrush are these changes on the EPC group's radar?