Meeting minutes
Extending the FHIR RDF ontology
Issue w3c/
ken: Eric Jahn and I have hosted some files on github.
EricJahn: https://
EricJahn: Ken started a unified care model.
… (UCM). And there's a requirements file.
… I forked the HCLS repo, and did a PR for my files.
ken: the dark blue are base classes in SULO.
… Discussed last week to use SULO as a foundation.
… Want a unified care model that goes across domains.
dbooth: How would the UCM fit with the FHIR RDF ontology?
ken: In addition to links to BFO, it will also link to FHIR RDF ont.
michel: I'm on my mobile, but will look offline later. Want to walk me through some of the classifications?
… Two design patterns: Process role, and SOLID pattern.
To what extent is the care portion aligned? How aligned to SULO?
ken: Basis: we created a new model where all of those classes and object properties were imported.
dbooth: Which classes/properties were imported?
ken: THe classes/properties from the SULO, described in the SULO paper.
… Then extending them in some ways.
… Into clinical or social processes.
… Person is a subclass of PhysicalEntity, which is a subclass of SpatialEntity
… I'm just playing around. Only the base SULO stuff is core.
dbooth: And is this the right way to use SULO?
michel: Base categorization is important, but also needs to reflect what the real world sees.
… Have you instantiated into anything yet?
ken: No. This group worked on FHIR RDF ont, and associated with records and clinical domain.
… Eric Jahn as worked on housing (from HUD model). Goal of the care model is to able to extend it into various domains, and classes/properties become domain-specific.
michel: I see a lot of good things, esp InformationEntity.
… Not sure what Participation class is. The name is a bit misleading, because it's an information object.
… You also have Report under Collection. Rationale?
ken: a Report is a collectoin of Observation results, and make have some additional characteristics. I'm experimenting with it.
… But I'm trying to understand what is an aggregation. A collection of statements?
michel: A Collection is much more like a set or class. You can have arbitrary members, like a stamp collection.
… But when you have a report, you have various parts.
… The things that are spatially-temporarily together are probably in the same space.
… Report seems more like an InfoEntity.
ken: Want to differentiate btwn what is an ont representation vs an epistemic representation. A report is an epistemic set if onstances.
dbooth: clarify? example?
ken: Eg, i've got a blob that captures some info about a condition.
… What could be captured in that is an ont model.
… But someone would make a statement about exactly what they did capture. And it would have values.
… A Report sometimes, like a med statement in FHIR, would be a collection about a specific med over time.
… Want to keep those separate.
michel: This is an area of extension in SULO.
… Maybe description, or argumentation, or counter arguments, descriptions of things.
… That just falls under InfoEntity, but isn't detailed further.
… Collection is more specialized.
ken: The driving business practice is taht in FHIR you get info from multiple systems, aggregating them, etc. Want to make sure that's a real value vs a conceptual value.
dbooth: Model level vs instance data level? Ken: Yes.
michel: Those both fall under InfoEntity.
ACTION: Michel to look at Ken and Eric Jahn's draft ont.
michel: Key part of SULO: the ont should be expressed w only the relations and no more. You replace yuour existing relations with SULO relations.
… Can you express everything in your ont with only SULO relations?
… Are there relations for which we have no compatible relationship?
… Is it fundamentally not expressible using SULO?
erip: Is there a LINT, for SULO?
… We express our ont, then ground it back to SULO. Is there anything tool that says "You missed this part"?
michel: SULOizer is trying to put a given ont into SULO.
… Big part of the challenge is to get at what the domain modeller was trying to do. It might not be evident from the name.
… Did it for FOAF and a couple others, but it will be a challenge. The idea would be to use an LLM to understand the categories and relations, to better understand.
… When we have a relation like hasMedicalOutcome, did we capture that?
… We were trying to describe equivalence using Shexmap
… Some kind of mapping language to say that elements on this side are equivalent to elements on that side.
… This is in line w the semantic interop.
ericp: What bits of the ont are not grounded in SULO?
michel: In FOAF there's an Agent class, which is a union of org or person -- completely different kinds of things. Doesn't fit cleanly w SULO. My SULOizer fails on it. Maybe they're roles.
ericp: But just having a punchlist of things that haven't yet been mapped back to SULO.
… There will be lots of cases in our world where we'll have the person/agent issue --- people or device groups.
ken: Yes, Participant. Org or person.
michel: Similar concept that you're asking. It says here are all the data properties, etc., but you haven't done these parts.
ericp: With shexmap, you can see what parts of the graph were not touched.
michel: I implemented your whole shexmap into suloizer. You can do it in the browser, then save it.
… And share it.
… I'll share it when it's stable.
dbooth: Sounds very analogous to the formal definition of semantic interop that I've been working on.
tim: “Two systems are semantically interoperable if […] all implications made by one system hold and are provable by the other, and that there is a logical equivalence between those implications” https://
tim: We also did a SPARQL query to check the mapping coverage. Does every term from one ont have some mapping to the other?
… And if you make it bidirectional, it becomes synonymy.
… I did a SPARQL query that every FHIR class had a mapping to W5.
michel: Maybe I can demo shexmap repo idea next week.
ken: How best to collaborate on this?
dbooth: Great to start writing things down, so that others can react to them.
tim: A few specific examples are really helpful. Even some mappings in spreadsheets.
ken: Still coming up with specific use case examples. But we have two existing ontologies: FHIR RDF ont, and the HUD model.
… Could we link those, so the UCM can support those existing models?
ken: I think we're a couple weeks away from an exmaple use case that uses them both.
tim: Keep in mind: FHIR has a way to make links to other terminologies within FHIR.
… We've worked in it the RDF way, with Concept IRIs.
ADJOURNED