Add checks to ensure has-soma-location is consistently populated (or consider simplifying the ontology)

[cmungall]

There are 54 asserted part-ofs between neurons and anatomical entities

(to get this list, use OAK to do this: cl relationships -p p .desc//p=i neuron)

These implicitly imply has-soma-location, but there is no way to express this in OWL, so we need an external rule to populate this

We also have the converse issue, in which has-soma-location is weaker, but many equivalence axioms and queries use part-of, which will not be satisfied in cases where it is "obvious" they should be. An example is the definition of "retinal cell" which uses part-of and is hence under-classified. See #2245

We may also want to reconsider the distinction. We could simply use part-of universally, and consider there to be a fuzzy boundary between the extent of projecting neurons on the one hand, and microscopic projections of macro brain regions on the other.

[balhoff]

If you add rolification axioms for 'neuron' and 'anatomical entity', and the property chain neuron o 'part of' o 'anatomical entity' -> 'has soma location', using Whelk you can increase subclasses of 'has soma location' some 'anatomical entity' from 275 to 517.

[cmungall]

nice! And I assume this means we'd get the relationships in RG too

We should also do:

hasPart o isNeuron -> inverse(hasSomaLocation)

(I am not sure this would yield inferences with whelk/elk due to the need for inverses)

Before taking this to the RO tracker, we should consider what our strategy should be

1. simplify (collapse relations)
2. upstream rule (e.g. sparql/dosdps) and assertion of redundancy
3. rolification
4. Both 2+3

I like 1 but I may be in the minority. I think 2 is the next most simple. 3 is elegant but could potentially lead to lots of confusion (e.g. I believe it's SOP for cl-editors to use elk not whelk). 4 is more robust but it can be confusing having multiple ways of doing things

[balhoff]

We should also do:

hasPart o isNeuron -> inverse(hasSomaLocation)

(I am not sure this would yield inferences with whelk/elk due to the need for inverses)

Nope.

[dosumis]

There are 54 asserted part-ofs between neurons and anatomical entities

(to get this list, use OAK to do this: cl relationships -p p .desc//p=i neuron)

What's the full OAK query for this?

I wonder whether some are just legacy.

We also have the converse issue, in which has-soma-location is weaker, but many equivalence axioms and queries use part-of, which will not be satisfied in cases where it is "obvious" they should be. An example is the definition of "retinal cell" which uses part-of and is hence under-classified.

I'm dubious of these grouping terms that are cover all cell types in some specific location - some users (e.g. CxG) really dislike them. With more specific genus terms, terms that group by location are not a problem (all neuron terms should use has_soma_location).

We may also want to reconsider the distinction. We could simply use part-of universally, and consider there to be a fuzzy boundary between the extent of projecting neurons on the one hand, and microscopic projections of macro brain regions on the other.

Please don't. Combined with other subproperties of overlaps in use for neurons, this would create a real mess for spatial reasoning. It would basically break VFB.

I like Jim's rolification suggestion. Should we add axioms to RO? (OTOH - another way to fix this might be to have a clearly defined standard for defining neurons - that all should have some part_of assertion associated with them. In most cases the part relationships will be to a much larger structure. We try to enforce this for FBbt (at least we have done in the past). We could even add a check for it.)

[github-actions]

This issue has not seen any activity in the past 6 months; it will be closed automatically in one year from now if no action is taken.