Semantic Web Health Care and Life Sciences Interest Group Teleconference -- 04 Mar 2015

<trackbot> Date: 04 March 2015

Topics for today

Eric: They're on this wiki: fhir-rdf XML->Turtle
... this wiki: https://github.com/w3c/hcls/wiki
... Want to transmute data from one form to another. How to take an obs, lab report and diagnosis and turn them from one form to another?

Dale: There are sister projects in HL7 to the CCDA on FHIR: "CDA on FHIR", which attempts to establish mappings between CDA r2 to FHIR (infra elements, authors, provenance, org, etc), whereas "CCDA on FHIR" is primarily about exposing within the CDA framework.
... It assumes that CDA on FHIR already exists.
... Want to incorporate the notion of CDA on FHIR for that?

Eric: I've done CCDA to RDF munging. Also done transforming of ORIM RDF. Also done transforming ORIM RDF to FHIR RDF.
... Former is XSLT; the latter is exemplary of turning FHIR RDF to FHIR XML.
... We can decide how much CCDA stuff is useful to this. This XSLT is hard coded: it doesn't come from the RMIM description, so it would be difficult to fit into any other notion of the RMIM.

<Alejandra> Would a generic tool like https://github.com/srdc/ontmalizer be useful?

Dale: CDA itself is an RMIM. CCDA is a profile (restriction) on CDA.

FHIR XML to Turtle

Eric: There's a github repo called w3c/hcls, and it is a parent of other repos.

https://github.com/w3c/hcls-fhir-rdf

Eric: Do a git clone of this repo, to start playing with it.

David: cd whatever-dir-you-want

<Marc_Twagirumukiza> https://github.com/srdc/ontmalizer was used in SALUS project where we were partner,... it seems to be handy than hard coded XSLT

David: git clone https://github.com/w3c/hcls-fhir-rdf.git

<Alejandra> See this too: http://www.srdc.com.tr/projects/salus/blog/?p=189

eric: ontmalizer is one mapper from XML to RDF. TBC has another.

Erich: Knowledge Explorer also has one.

Eric: Goal of this one is peculiar to what we want to get out of FHIR XML as RDF -- much more custom, and produce an RDF that we want.
... It's possible that these other tools will produce an RDF that we want.

Tony: Is this instance or model transformation? TBC is model transformation.

<Alejandra> https://github.com/w3c/hcls-fhir-rdf/tree/master/generic

<ericP> https://github.com/w3c/hcls-fhir-rdf/blob/master/generic/transform.xsl

Eric: In that dir there's transform.xsl, and it's a big chunk of XML.
... Line 19 is l:fhirdefs . There's a python program that reads the FHIR definition, then concatenates an XSLT head, tail and body to product transform.xsl

<ericP> https://github.com/w3c/hcls-fhir-rdf/blob/master/generic/transform.xsl#L19381

Eric: See line 19381 . Below that is the custom stuff that makes the RDF that comes out more palatable.
... This overrides the generic transform, including ATOM, for putting multiple things in one doc.

<ericP> https://github.com/w3c/hcls-fhir-rdf/blob/master/generic/transform.xsl#L19575

lloyd: Before we jump to transforms, start from an example instance, and say "this is what we want the RDF to look like", and then craft the docs to say "here's how this FHIR data is represented in RDF".
... Then once we have that consensus, we do the grunt work of writing transforms, updating reference implementations, etc.

Eric: This is the process that we went through before. Then we tempered that with what's intuitive to FHIR developers and what's manageable to code. But we're in a new community now, so it makes sense to repeat that process. (Previously Josh and me.)
... Prefer to start with what we already did.

<Zakim> dbooth, you wanted to ask about first doing a totally generic translation, and then SPARQL transformation to produce better RDF.

<ericP> dbooth: could start with generic translation from XML->RDF | RDF->RDF using SPARQL or ShEx

Tony: Agree with lloyd. First agree on an example, so we're not jumping around between different resources. Second that's different from what happend before is we're looking both at model transform and instance transform, and i think we need to start with model transform first, or else instance won't match it.

<ericP> examples that we used before.

Tony: Then we can do the coding to do the model and instance transforms.

eric: you want to look, in a green field, what you'd like, and then build the transformation?

tony: yes. If what you've done doesn't fit with the model transform, then it won't work.

<Lloyd> Example requirements: nested structures, complex data types, extensions, id references, narrative, choice data types, recursion, contained resources, multiple coding repetitions, fixed and preferred bindings

marc: Need to draft the vocab we want -- FHIR ont. Coming before or after?

<Zakim> dbooth, you wanted to say I think we should first see what Eric and Josh did, then we can do greefield if we want.

dbooth: I think we should first see what Eric and Josh did, then we can do greefield if we want.

lloyd: I posted in IRC the characteristics I think we want.

tony: Probably want local (HL7 FHIR) and remote bindings too.

<rhausam> my call dropped and the conference bridge won't let me back in - says call is restricted

lloyd: Separate requirements from a binding perspective

rhausam, try again

<rhausam> still "conference is restricted at this time"

<ericP> starting examples

rhausam, sorry cannot fix it now. try to follow in IRC if you can.

<rhausam> ok

Eric: That link is what we tried to address.
... The XML docs were the starting points, the RDF docs were the result.
... That was the Turtle that we said we wanted for those exmaples, and then we crafted the code to make them happen.
... But you might want to do them differently.
... Re model versus instance first, that depends on whether you're a top-down or bottom-up person. I prefer instance first, then gen model after I like the instances.
... The cost of that is that sometimes you can write yourself into a corner, but probably rare.

<Alejandra> I will have to leave now - bye everyone!

Eric: Dividing the schema into unambiguous space (avoiding same URI for two attributes with the same name in different resources, which may have different meaning)
... The default is they're all distinct unless we say otherwise.

tony: Is your work separated between the model and instance? Seems a combination of the two.

eric: In RDF when you define instance you have a model.

lloyd: Can infer model from the instance provided you have knowledge of the resource.

eric: Same python tool could spit out OWL.

<Marc_Twagirumukiza> I would support first to work on the model ---f.eg using FHIR resources as model

lloyd: Shouldn't directly mention dc: terms in instance data. That should be in the ont.

<Lloyd> https://github.com/w3c/hcls-fhir-rdf/blob/master/generic/tests/diagnosticreport_list-thin.ttl#L27

dbooth: agree

<Marc_Twagirumukiza> +1 I think we will need to work both on the ontology and on the model before we jump on instances

eric: Pushback: When possible, good to go directly to an ont that others understand. Maybe at this point there's nothing to come from popular RDF onts.

Marc: I think we will need to work both on the ontology and on the model before we jump on instances
... because when we go to instances, we need to link instance data to link to vocab and model. Need to work on both, then instances.

eric: It's a detail that we can figure out later.

guoqian: When we transform FHIR to RDF, have we discussed unique URIs for FHIR elements? Different implementations might use different namespaces. Need a common ont in order to ref std FHIR.

Eric: I wrote into HL7 namespaces all of the predicates and classes in these examples. Those imply a schema that would probably be hosted at that URL. Those what you mean?

<Marc_Twagirumukiza> there are some benefit to use a single namespace fhir:

Guoqian: yes, but I've seen others that use other namespaces.

<Marc_Twagirumukiza> because this can help to make reusable predicates at ontology (vocab) level

lloyd: I'm comfortable using those as the prefixes, then once we get the web site doing what you want, if you queried that URL with appropriate request header, you'd receive the appropriate Turtle, OWL, etc.
... Conneg.
... it will be .../fhir/structuredefinition/...

eric: We can patch that stuff up later.
... Who will check this out of github, and make test and try changing some things?

<DNelson> yes I'll try

<Ingeborg> planning to try

dbooth

<Ingeborg> Ingeborg

<Marc_Twagirumukiza> Marc

<egombocz_> planning to try

<Marc_Twagirumukiza> am on windows

<Marc_Twagirumukiza> i will explore how

<Lloyd> Lloyd: won't do make, but will review and identify the things I think should change/need to be added

eric: preferred list?

dbooth: Suggest both HCLS and ITS lists unless we start annoying people with too many messages.

tony: choose a sample first?

http://wiki.hl7.org/index.php?title=FHIR_Examples_For_Driving_FHIR_Ontology_Development

<Zakim> dbooth, you wanted to ask about dependencies

dbooth: dependencies?

eric: when you do "make site" it pulls down stuff that's needed.
... Two things we want to grab from the FHIR spec. One is the XSLT. Another is samples from the FHIR spec, though currently not, because they were heaviliy edited to reduce their size.

Dale: Notion of an example, I suggest rather than trimming stuff out, grab the simplest we can. There's a lot of chatter about how to do things like problem lists. Lots of discourse around them still. Suggest something that is well establish in FHIR and simple enough, without removing material for our purposes.

ADJOURNED

- DRAFT -

Semantic Web Health Care and Life Sciences Interest Group Teleconference

04 Mar 2015

Attendees

Contents

Topics for today

FHIR XML to Turtle

Summary of Action Items

Scribe.perl diagnostic output