Create new UI for hinting evidence with GP individuals

[kltm]

This is the Noctua component of the Minerva item geneontology/minerva#99.

We would like to add a mechanism so that optionally, when adding evidence, an additional virtual edge annotation can be added for the relevant GP in the model.

To distinguish different individuals of the same ID, fruit icons (or similar) could be used.

It would likely be implemented (if done before a larger refactor) like the evidence cloner.

Depending on how deep the necessary changes are, #141 might come into play (as well as extended testing, etc.)

[balhoff]

@cmungall any thoughts on property to use? What about http://purl.obolibrary.org/obo/RO_0002472 is evidence for?

[kltm]

Also note in #141 the use of http://purl.obolibrary.org/obo/RO_0002162 axiom has evidence.

[cmungall]

AP or OP?

[balhoff]

I was imagining it as OP from evidence instance to gene product instance.

[kltm]

Okay, so it seems that @cmungall would be fine either way.

As OP, without resolving #141 or making any changes to the client, the effect would be unpleasant (I just tested it). Evidence has always been a special case--guaranteed to be singletons--and having them connected is not dealt with well.

So, I think the question here is then how much work do we want to put in before progressing? While I'd just love to just tackle #141, and fix this at the same time, I'm realistically wrapped up until the pipeline transfer is completed, and that has been slow going.

The "on the cheap" approach of AP might make quicker progress. AFAIK, the only place we've used something like that so far is in the modeling of the evidence, creating an "edge" from an axiom. With my understanding of what is currently going on, I think that would be a fairly safe and "invisible" change to the client.

Any thoughts on the effort involved at your end, or possibilities of future modeling issues?

[kltm]

@balhoff @cmungall @goodb I just talked to @thomaspd and was interested in how we should move forward on this. Also see geneontology/minerva#99. I think it's important to get the most "correct" modelling solution, but realistically some will require more work than others to get into place. Would there be any issue with pursuing the AP solution; are there any planned extensions to the model that this might be disruptive with?

[goodb]

@kltm I hesitate at introducing the concept of a special 'virtual edge' just for the UI. If I understand correctly from this and #99 , it sounds like these are edges that would be inferred by the reasoner, have not been directly asserted in the model by a person, are of interest to the curator, and the curator may have additional evidence to assign to them. From a computational perspective, these edges are real. They will appear in the RDF view that we will be accessing from the SPARQL endpoint. From the human perspective they are also real as the curator user has something they want to say about them.

Based on that. I think the most straightforward solution here is to enable the curator to add the edge to their model and add evidence as they see fit. This could be enabled in the UI by (1) running reasoner, (2) showing inferred edges like these, (3) allowing the curator to click on them and open the evidence assertion view as they normally would on other edges and (4) if the curator wants, store the assertion and evidence like any other claim in the model.

@balhoff When there is no additional evidence to add beyond the fact that an edge was inferred, this should likely be captured in an Evidence item structured just like the others such that the information is present for SPARQL query. Is something like that part of the pipeline for building the rdf endpoint?

[balhoff]

@goodb the procedure you're describing is similar to an approach I advocated at one point. Especially because in the current system the evidences aren't even really on the right edges—very often the evidence does not support the actual semantic relation being annotated, but instead supports the GPAD annotation which will ultimately be exported. I feel like the UI could list in a table all the direct GP-to-GO relations which are being inferred from the model, and attach evidence to the ones they want to show up in GPAD.

[goodb]

@balhoff "very often the evidence does not support the actual semantic relation being annotated" sounds like a serious problem. I would strongly advocate for keeping the OWL model coherent in itself as priority number 1 and isolating code to support the production of GPAD or other views on that model.

[kltm]

I think we may have to re-prosecute this. Although I think we were all convinced with the "right" way forward after the meeting, it is far enough back that we seem to have lost track of the reasoning and the next action. It would be good to get a call together including @ukemi @vanaukenk @cmungall to hash this out again.

[vanaukenk]

"It would be good to get a call together including @ukemi @vanaukenk @cmungall to hash this out again." Yes, that sounds like a good idea.