Closed cmungall closed 2 years ago
There is a general pattern here of a TBox classification mirroring an ABox tree, where the ABox tree represents in some sense prototypes, exemplars, canonical instances, etc.
We have two object properties in the following patterns connecting two bipartitle graphs
This is broadly useful for a number of common scenarios: (1) real world biology is often messy and variable, but ontology TBoxes work best with invariant assumptions (see Rector, Solbrig et al). This allows for canonical examples in the abox that correspond to a domain scientists conceptualization of a "typical" case, and an intuitive mirroring in the abox that does not make incorrect all-some assertions (see Schulz et al) (2) data-driven approaches including but not limited to sc data provides arguably more biologically relevant classifications (via clustering methods etc), combining these with top-down classical classification can be very powerful
This can also be useful for avoiding authoring duplicative assertions when using existential axioms where the axioms hold in both directions
[table below still being fleshed out]
use case | has prototype relation | ABox domain | subtyping relation | cite |
---|---|---|---|---|
cell classification from scSeq | has exemplar | clusters | subcluser_of | ... |
canonical anatomy | has canonical example | anatomy instances | subexample_of | ... |
ontology recapitulates phylogeny (genes/proteins) | has sequence | sequence trees | evolved_from | ... |
ontology recapitulates phylogeny (anatomy) | has prototype | evolutionary prototypes/types | evolved_from | ... |
Sorry ... I'm not following how you are distinguishing between prototypes and exemplars in your examples.
In common usage exemplar seems close to what you might call a canonical individual. But, prototype is a model of something (i.e., not canonical).
I'm not opposed to the has exemplar
OP, just trying to understand the intent better.
I suppose I’ll jump in here with my thoughts: I think this is a good point, @wdduncan, and some clarification would help. I didn’t think of prototype
in the way described by your link, but closer to the lexical semantic sense (Rosch). Similarly, I thought of the lexical semantic sense of exemplar
(Nosofsky).
These senses would (very roughly) be something like:
prototype
: a single instance of a member of a category that is an abstract ideal.
exemplar
: an actual member(s) of a category that is typically representative of members of that category.
In category theory (from what I understand, which is not much), prototypes are typically abstract entities and exemplars are typically concrete. There is almost always only one prototype for a category and typically more than one exemplar for a category. I’m sure there are other differences (again, not an expert by any means), but I don’t know they are relevant here.
So, I think the has_exemplar
relation is correct for the scSeq example when thought of in this way. Thoughts?
@patrick-lloyd-ray Thanks for the clarification of the different senses for prototype
and exemplar
used in category theory. I was not aware of these :)
So, in these senses, would a reference genome be an exemplar (assuming the reference genome is a concrete individual)?
closer to the lexical semantic sense (Rosch). Similarly, I thought of the lexical semantic sense of
exemplar
(Nosofsky).
@patrick-lloyd-ray Could you provide more detail for the references to "Rosch", "Nosofsky" and category theory (URIs or bibliographic detail)? Thanks!
@wdduncan It seems so, although I'd defer to @cmungall on this. I'm not sure that what I've said was intended by the proposal, but your question was a good one and made me think about how one should approach this.
@cboelling No problem! I think the most cited are: Rosch, E. (1975). Cognitive Representations of Semantic Categories. Journal of Experimental Psychology: General 104: 192-232.
Rosch, E. and C.B. Mervis. (1975). Family Resemblances: Studies in the Internal Structure of Categories. Cognitive Psychology 7: 573-605.
Nosofsky, R.M. (1988). Exemplar-Based Accounts of Relations between Classification, Recognition, and Typicality. Journal of Experimental Psychology: Learning, Memory and Cognition 14: 700-708.
Without endorsing the work as a whole or the conclusions contained therein, I think that Lakoff has a nice and short historical summary of category theory in the opening chapters of "Women, Fire, and Dangerous Things" (ISBN:0-226-46804-6).
Added terms in closed PR https://github.com/oborel/obo-relations/pull/503 -> there are some discussions there that might need continuation and refinement of the term but sufficient for now and we do want to pull it in for BDS ontology.
so can this ticket be closed?
yes, sorry, totally forgot to close it
We need a relation 'has exemplar' with a particular use for the cell ontology (and the brain data standards ontology) to link a cell type to a cluster from sc transcriptomics. The general pattern may be useful elsewhere.
For the clustering use case we may want relations like subcluster_of (taking RO into information artefacts, where haven't gone so far), so we can define property-chains such as 'has exemplar' <- 'has exemplar' o subcluster_of
This is closely related to an existing relation 'has prototype' http://purl.obolibrary.org/obo/RO_0002214