Meeting minutes
Concept IRIs
gaurav: Added text for both flat and hierarchical IRIs.
… How do we convince other people of the value of this?
… Still haven't found a real life hierarchical example.
… After looking at the document, the example is only flat IRIs.
… We should update to show hierarchical example.
dbooth: What should discussing this a next week's HL7 meeting?
gaurav: Concern if someone looks at this in incomplete state and gets a bad impression.
dbooth: I think it would be better to use a few slides than showing the document. THe doc looks too messy.
ACTION: DBooth to draft slides outlining the concept IRI proposal
dbooth: Opinions about a flat and hierarchical approach.
eric: I favor the dual flat and hierarchical approach.
gaurav: Leaning toward template idea, then flat only, then flat+hierarchical.
… We don't have a good example of a sys that would benefit from hier approach. Ontologies often have multiple hierarchicals.
… Also would be better to get flat concept IRI first, then add hier later if needed.
eric: How about saying SNOMED has 400K+ terms, but they could be more hierarchical. There's a classification that says whether something is a measurement. If people were reworking SNOMED, this is how they could do it.
dbooth: I feel fairly confident that FHIR folks will push back against features for which there is not a known need.
eric: Potential for having real web spaces for these terminologies.
jim: Leaning toward keeping it simpler, esp because we don't have an example from the wild that needs hierarchical.
ACTION: Gaurav to send DBooth template-based example
jim: When you write a Human Phenotype Ont identifier, like HP:0001671
… When that's turned into concept IRI, the "HP:" part gets stripped off of the code before appending to a stem IRI
eric: Algo would need a list of those prefixes that are in scope. Would require regex hacking or prefix definitions.
RDF lists
jim: Made more progress. Opened an issue at OWL API. I now understand what happens. Parser encounters a reified axiom, w subject a bnode, those refer to the triples of the axiom being annotated. It hands them to the different kinds of tirple handlers. But just before it does that, it gets the annotations.
… Some turtle handlers don't make an axiom or take pending annotations, so whatever comes next gets them.
… If I had a way to clear the pending annotations, that may solve the problem. Goal is to open a PR to fix the issue.
… If I fix the issue, we could use RDF lists as we're talking. But might not be easy to retrieve axioms.
dbooth: ETA to get it resolved?
jim: A month?
… I'm not going anywhere this summer.
eric: I think it's a great bug to fix in OWL anyway, and I'd really like to see us use RDF lists if possible.
… Didn't check whether it screws any of our use cases.
jim: THere's an issue w classification use case, unrelated to parsing. List items must be anonymous individuals, bnodes. If you're using SNOMED, you'll have to use probably Elk or SnoRocket. OWL EL profile does not support anon individuals.
eric: I saw it work prototyping this, I think using hermit.
jim: Hermit will work, but the issue is scalability. SNOMED is too big for it.
eric: I hand picked the subset of SNOMED that I needed.
jim: triple store reasoning like Jena rules won't care.
gaurav: Seems like a bug worth fixing in OWL. Could fall back to R4 approach if we don't fix this in time.
FHIR RDF Playground
https://
eric: Implemented property shortening, for both resource and datatype (using toggles on top left).
eric: You can also load a different set of structure defs.
ADJOURNED