Reasoner broken in noctua?

[dosumis]

I expected to see some inference on this model:

Including these inferences on the cell proliferation node:

Any clue why this is not working?

[kltm]

The reasoner seems to be working as far as the stack goes. Maybe @cmungall or @balhoff could speak to why you're not seeing what you expect here.

[cmungall]

Unfortunately the configuration is not very introspectable, but I suspect it's because minerva is using --slme-elk. The deeper classes will typically not be in the module.

We experimented with just using elk, no module extraction. I hypothesized this would actually be much faster since most of the slowness was actually in the module extraction step. AFAICR we were still somewhat reluctant to switch the config, but I don't recall why

[cmungall]

Previous discussion here: https://github.com/geneontology/minerva/issues/33

And yes, use of slme-elk is hardcoded in the gulpfile:

https://github.com/geneontology/noctua/blob/master/gulpfile.js#L390-L406

[balhoff]

I checked and slme-elk is in use in the deployed Minerva. However I loaded that file into Protege and after running ELK I don't see the inferred types that @dosumis showed. @dosumis are you using only official release ontologies?

[balhoff]

The model has an instance of CL:0002405 gamma-delta thymocyte which doesn't have any semantics in go-lego.owl. gamma-delta T cell proliferation GO:0046630 refers to gamma-delta T cell CL:0000798 in its logical definition.

[cmungall]

@balhoff if you have the same abox model combined with go-lego, and the proliferation class is not deepened, it could be because

• Elk is not complete for Aboxes (still not sure which cases it covers and which it does not)
• Different DP and/or OP is being used

[balhoff]

This works if you additionally import the complete cl.owl, but not with go-lego.owl. @dosumis had you added an import of CL?

[dosumis]

This works if you additionally import the complete cl.owl, but not with go-lego.owl. @dosumis had you added an import of CL?

Yep.

The inference to regulation of proliferation should work without this though. I’ve seen work it in other models.

[dosumis]

Similar inference on reg seems to be working fine here:

http://noctua.berkeleybop.org/editor/graph/gomodel:580685bd00000071

[balhoff]

The specific problem we have is that the axiom gamma-delta thymocyte SubClassOf gamma-delta T cell is not in go-plus (i.e not in cl_import). I guess because gamma-delta T cell is referenced when making the module but all its subclasses are not pulled in. If you can reference any CL or UBERON term within Noctua models, reasoning will always be potentially incomplete if the full versions of these ontologies aren't included.

[dosumis]

If you can reference any CL or UBERON term within Noctua models, reasoning will always be potentially incomplete if the full versions of these ontologies aren't included.

Would it be viable to run a reasoner with complete ontologies (rather than just imports). If not, can we run SLME in the background to generate ontology slices on the fly for use in reasoning. Perhaps this could be run in the background as terms are added?

BTW - doesn't TermGenie cope with this issue, and if so, how?

[kltm]

Any comment on the current status of this ticket? If it is still open/ongoing, please change the title to something more specific.