HCLSIG/SWANSIOC/Meetings/2009-03-20 Conference Call

Conference Details

Date of Call: Friday March 20, 2009
Time of Call: 11:00am Eastern Time
Dial-In #: +1.617.761.6200 (Cambridge, MA)
Dial-In #: +33.4.89.06.34.99 (Nice, France)
Dial-In #: +44.117.370.6152 (Bristol, UK)
Participant Access Code: 4257 ("HCLS")
IRC Channel: irc.w3.org port 6665 channel #HCLS (see W3C IRC page for details, or see Web IRC)
Duration: ~1 hour
Convener: Tim Clark
Scribe: Harold

Agenda

Featured John Madden, MD, Duke University will present on terminologies vs ontologies in scientific discourse. A great opportunity to discuss SKOS vs OWL and their most appropriate use cases. Slides
Sudeshna to give a report on the status of SCF semantic microarray repository
Participants to review peoples' takes on the SWAN-myExperiment integration path

Minutes

Attendees: Alex, Peter, Paolo, Kei, EricP, Holger, David, Ronald, Marco, Carl, Harold, Elizabeth, Scott, John, Susie

<ericP> topic: JohnM presents on terminologies
<ericP> Term preso http://tinyurl.com/cssred
<ericP> ... i gave this preso at C-SHALS
<ericP> ... timClark thought it was relevent to scientific discourse
<ericP> ... he was trying to get scientists to formalize there knowledge in OWL
<solbrig> Tim had issues with scientific discourse ecosystem, but found that scientists didn't speak OWL
<ericP> ... came to the conclusion that scientists don't speak owl
<solbrig> Terminologies don't really speak OWL either
<solbrig> You are not crazy when you can't put terminologies in OWL - it is actually hard to do.
<ericP> johnM: slide 2 represents the state of the art in terminologies today
<ericP> ... e.g. lab results use LOINC...
<solbrig> johnM: some are just term lists
<solbrig> ... some are hierarchical
<solbrig> ... documents will be encoded in more than one vocabulary
<solbrig> ... what is unusual to modify CPT code w/ LOINC or SNOMED w/ HL7
<solbrig> johnM: slide 3 - terminologies came first
<solbrig> ... semantic web, syntax came first then content.
<solbrig> ... vocabularies are inherently mixable on the web
<solbrig> ... central curation is not impossible, but you can't prevent them from using your terms in their assertions
<solbrig> ... SNOMED is the giant among vocab - 1.25 million terms, web will probably be < 200 entries per vocabulary
<solbrig> ... and will be very intensively modeled, highly contextualized
<solbrig> johnM: slide 4 - 3 challenges
<solbrig> ... common identifiers
<solbrig> ... keeping faith with original intention of the vocabulary
<solbrig> ... using GLUE technologies
<solbrig> johnM: Slide5 - common identifiers
<solbrig> ... no problem coining URL identifiers for terms. It really isn't our place, really the place of the central curation authority
<solbrig> ... existing software doesn't expect URL's
<solbrig> johnM: slide 7
<solbrig> ... identifiers for individuals. I had a a hernia sack repaired, it is John Madden's hernia sack.
<solbrig> ... existing terminologic systems - subject of assertion is implicit, not explicit...
<solbrig> ... documents encoded in SNOMED - just a list at the bottom of the document. HL7 CDA - same way, still hard to code in triples
<solbrig> ... (rdf square brackets in slides mean "instance of")
<solbrig> ... really trying to encode "surgeon reports a radiological history of lung mass" vs. "lung mass exists"
<solbrig> ... existing terminologies don't worry about this because it is hard to do, but in RDF you've got to
<solbrig> ... would like a very small, compact vocabulary of assertions (FOAF of DC for medicine)
<solbrig> johnM: slide 8
<solbrig> ... Slide 8, 9, 10, 11, 12 13 are use cases at varying levels of inference
<solbrig> ... slide 17 - links. No ontology needed, all there is is links (lhs of side). RHS is examples of intensive inferencing
<solbrig> ... all elements of slide are wonderful Semantic Web use cases. We aren't failing if we don't deliver the "grand OWL ontology of medicine"
<solbrig> johnM: slide 18
<solbrig> ... a lot of the semantic web will links between heavily modeled, narrow ontologies...
<solbrig> ... experts will have sw equivalent of web site - ontology of rheumenologic (sp?) diseases. You will have to use his knowledge from the web site to do inferencing
<solbrig> ... grab individual expert sites and apply to your problem.
<solbrig> johnM: slide 20
<solbrig> ... scientists will formalize tentative ideas. Can swap ontologies in and out to perform interesting.
<solbrig> ... it is not interesting to know that influenza is an infectuous disease.
<solbrig> Peter Hendler: We are moving SNOMED CT into OWL, and will be presenting in IHTSDO in the first week of April.
<solbrig> Peter Hendler: Will be able to share it the first week of April.
<solbrig> .... took SNOMED Tables, used Spackman's conversion script to OWL and then added 25K negated terms.
<solbrig> ... will be adding incrementally negation and disjunction as needed.
<solbrig> johnM: slide 21 - from SKOS. Notion of "features"
<solbrig> ... using SKOS (vocabulary expressed in OWL, for describing terminologies, no assumptions about model theory)
<solbrig> ... allows to get knowledge from terminology without forcing it into a particular model.
<solbrig> ... "meta-inferencable" vs. "immediately inferencable"
<solbrig> ... not what we might want it to mean, what it really means.
<david_newman> SKOS was made a W3C candidate representation sinc 17th March
<solbrig> ... open world principle. If it isn't said, it isn't necessarily false.
<solbrig> ... danger of adding your own interpretation vs. those of those who wrote the terminology.
<solbrig> ... in OWL, if you don't make statements about disjointness, they aren't assumed to be disjoint. UMLS Semantic Network example...
<solbrig> ... John added disjointness axioms, but Olivier Bodenreider (curator of SN) said it wasn't what he meant.
<solbrig> ... disjointness issues manifest on the instance level.
<solbrig> ... relationship overloading. "isA" is not just "isA" - could be RDFS:subclasOf, RDFS:subPropertyOf, RDF:type, equivalentClass, etc.
<solbrig> ... might inverse of owl oneOf, owl unionOf. Choice of meaning has consequences.
<solbrig> johnM: slide 28 - introduction to AIDA - SKOS Semantic Aggregation Environment
<solbrig> ... slide 29 - things you can say in SKOS
<solbrig> ... AIDA knows about SKOS and SKOS terminologies. Slide 30
<solbrig> ... SNOMED is just represented as BT and NT relations - hierarchies.
<solbrig> ... slide 32 shows grabbing RH middle pain of SNOMED hierarchy as a whole, left hand is lucene index of Medline
<solbrig> ... semantically enabled full text search. (Scott: index is just of Medline abstracts - not full text)
<solbrig> johnM: slide 33 - glue technologies
<solbrig> ... most of the artivacts of the medical world are "documents" - authorship, stewardship, atomized data, natural language
<solbrig> ... slide 35 - have additional characteristics beyond web
<solbrig> ... legal status, beginning and end, non-repudiation, default context.
<solbrig> johnM: Slide 36 - Plato's ?? translated into latin by someone else, with commentaries, commentaries on commentaries, etc. purchased by vatican
<solbrig> johnM: Slide 37 - ontologies have resemblance to forms, but more complicated.
<solbrig> ... documents get interpreted to create semantics. Semantics extracted using GRDDL - assigning intrepretation engine to a particular document.
<solbrig> ... semantics change, add a different GRDDL extraction algorithm to it
<solbrig> ... (slide 38)
<solbrig> ... corollary in scientific discourse as well. Significance of entities in paper isn't understood until other papers emerge. Active of interpretation brings knowledge to document
<solbrig> Susie: time left... questions?
<solbrig> Peter Hendler: will you be at IHTSDO in April?
<solbrig> johnM: no
<solbrig> Susie: HCLS FTF in Cambridge on April 30 and May 1. Invited talks on where group is, breakouts
<solbrig> johnM: won't be able to attend on Thurs, will be on Firday
<Susie> Agenda & registration for F2F meeting -> http://esw.w3.org/topic/HCLSIG/Meetings/2009-04-30_F2F
<solbrig> reference to Rector's paper in 2006 in model of talk? and model of meaning
<solbrig> Alex: Space of code? and space of meaning. This is why we picked SKOS vs. something else.
<solbrig> ... last version of SKOS is submission (W3C) proposing SKOS / OWL version ... (unintelligable)
<solbrig> ... why SKOS vs. OWL DL?
<solbrig> Alex: OWL Full collapses punning, OWL DL has two classes.
<solbrig> Alex: in OWL 2 that isn't true anymore.