See also: IRC log
<trackbot> Date: 25 August 2015
<dbooth> Scribe: rhausam
<ericP> en route
david: defer minutes until we have quorum - no chair yet
tony: presented to ARB last
Thursday on FHIR mapping to O-RIM
... issues with value set bindings - how do we declare the code
system and concepts?
... divergenge with FHIR
... in RDF/OWL code system is a namespace where a code is
unique
... code is a fragment of a URI - code system is a prefix
(?)
rob: code system is a "namespace" generally - not specific to FHIR
tony: code system would be a
named individual
... that is consistent with O-RIM
... concept may be a single class in RDF
rob: code system is a class
tony: agree
... something missing - may be part of the problem -
systemconcept class
david: can you run through that again?
tony: concept "red" - system concept would say that this is "red" as defined by SNOMED CT (for example)
david: think you are referring to in an observation when there is more than one code indicated - this observation is both of those system concepts
tony: intersection of both types
rob: but the two types are not necessarily equivalent
tony: correct
<dbooth> david: When an FHIR observation says that it is SNOMED code 123 and LOINC code 456, it is saying that the observation is both of those concepts -- a subclass of both of those concept classes.
rob: but "red" generally doesn't actually exist (outside of the code systems)
tony: do we declare the abstract concept? we might or might not
david: unless we tie into an upper ontology, there may not be any of these high-level declarations
tony: 3M NCID does this
rob: any number of choices could be made for the "upper" concepts - including SNOMED CT
david: I'm fundamentally skeptical about having any concepts that are not tied to system
tony: value set binds to system
concepts, not to the "abstract" concept
... relationships are brought to the object property level -
this is where the problem is, since object properties relate
individuals
... fundamentally different from FHIR and O-RIM - declare the
values to be a set of individuals
rob: choice of modeling - has consequences
tony: the question has been are concepts classes or instances?
rob: the answer is yes :)
tony: probably need both
david: discuss with Barry Smith in September
tony: don't know his position on this issue
<dbooth> david: You are modeling a valueset as a union of concept classes, because each concept is a class rather than an individual.
<dbooth> david: Whereas ORIM models a valueset as a set of individuals.
tony: need grounding and consensus in how to express concepts in RDF/OWL
tony; CTS deals with instances
<dbooth> david: Maybe we should discuss this question with Barry Smith at the CTS ontology workshop in September http://ncorwiki.buffalo.edu/index.php/CTS_Ontology_Workshop_2015
rob: not specified whether instances or classes in CTS
<dbooth> ISSUE: Should a ValueSet be modeled as a union of concept classes, or as a set of concept individuals?
<trackbot> Created ISSUE-15 - Should a valueset be modeled as a union of concept classes, or as a set of concept individuals?. Please complete additional details at <http://www.w3.org/2014/HCLS/track/issues/15/edit>.
tony: need to decide this
... SNOMED concept is an identity - label tells what it means -
is an OWL class
... FHIR value set definitions - object of the "include" is an
individual
david: tony, can you show an example?
tony: show SNOMED example
rob: SKOS represents as individuals
tony: SKOS fits at metamodel - individuals can be puns of classes
rob: have done this with SKOS
david: what other relationships and properties do you want to define?
<dbooth> rob: if you have the concept Red in two terminologies, it isn't safe to say that they are exactly the owl:sameAs, but with SKOS you can say narrower/broader.
tony: can also say things similar
to narrower, broader in OWL by subclassing
... mapping is "comfort food" - doesn't necessarily mean
anything
... "equivalent" concepts may break systems when substituted -
can be a dangerous thing
... if you find something that needs to be expressed beyond
equivalence, disjoint and subclass, can use SKOS and
punning
... proposing that concepts and value sets would all be
declared as classes
... instances of concepts are artifacts in the medical
record
... would be happy to do a paper on this (could be multiple
authors) - how to define code systems, value sets and
concepts
david: should discuss this with Barry Smith in September
rob: want concrete material to discuss with him
david: OWL full allows something
to be both an individual and a class - non-OWL-full likes to
keep classes and individuals as distinct
... sort of an artificial distinction between classes and
individuals - tooling issue
rob: nothing
tony: was getting confused on how O-RIM and FHIR were doing this, and what we want to do in RDF/OWL
<inserted> Scribe: dbooth
eric: Cecil wrote programs to
swap between forms of expressing valueset concepts (as classes
or as values).
... Tony
... Tony's suggestion (to model valueset concepts as classes)
is consistent with other things we have done so far, but we
should look at the expressivity that we'll need, and consider
whether we'll need them, versus modeling them as
individuals
david: Downside of concepts as classes?
eric: The system has to work
harder. You're adding more inferences. It's gotten in Cecil's
way in the past. Should ask Cecil where/when it got him into
trouble.
... I took apart gender in the ORIM and it took me ages.
tony: ORIM is very difficult to understand.
eric: another thing that would
simplify ORIM a ton would be to change 'act' relationships with
a type code, to that type property directly.
... E.g., if the type is 'severs', then use a 'severs' property
directly.
<scribe> ACTION: Tony to draft a paper describing how valuesets and values in a valueset are represented [recorded in http://www.w3.org/2015/08/25-hcls-minutes.html#action01]
<trackbot> Error finding 'Tony'. You can review and register nicknames at <http://www.w3.org/2014/HCLS/track/users>.
david: recap discussion for Eric who just joined
eric: two different patterns
affect what inferences you want to support, and how much "coal"
you need to supply
... what tony is proposing is more consistent with the other
decisions that we have been making, but we should (a) determine
if we need that expressivity and (b) what do we lose if we turn
that expressivity down?
tony: value set would be a union
of the concepts - not a superclass - that's different
... could declare the coding to be the value set
rob: don't think that's right
tony: that's unspecified
rob: right
tony: focus on what code systems, value sets and concepts are
david: write this up as a
separate write up - discuss with Barry in Charleston in
September
... downside of going with concepts as classes?
eric: makes the system work
harder
... ask Cecil at what level he ran into difficulty with this?
what use cases needed this expressivity?
tony: O-RIM is very complex to understand - wanted to try to simplify it - that's a major "no-no" for what we want to do here
eric: would really simplify the
O-RIM to remodel ActRelationship objects - replace them with
arcs
... effectively a reification
david: Google doc?
tony: prefer to send emails -
Word doc
... prefer to go with smaller group at first while draft, then
make public
... will try to have draft by this weekend
ADJOURNED
This is scribe.perl Revision: 1.140 of Date: 2014-11-06 18:16:30 Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: RRSAgent_Text_Format (score 1.00) Succeeded: s/SKOS/SKOS and punning/ Succeeded: s/while still keeping/non-OWL-full likes to keep/ Succeeded: i/swap between/Scribe: dbooth Found Scribe: rhausam Inferring ScribeNick: rhausam Found Scribe: dbooth Inferring ScribeNick: dbooth Scribes: rhausam, dbooth ScribeNicks: rhausam, dbooth Present: David_Booth Tony_Mallia Calvin_Beebe Darrell_Woelk Rob_Hausam EricP_(last_15_minutes) Found Date: 25 Aug 2015 Guessing minutes URL: http://www.w3.org/2015/08/25-hcls-minutes.html People with action items: tony[End of scribe.perl diagnostic output]