Re: Primer OWL Entailment

On Tue, 20 Mar 2007, Harry Halpin wrote:
> Chime,
>    If you can, remind me of one simple OWL entailment that would make
> sense with the clinical data example in the GRDDL Primer and I'll add
> the verbal description and check it into the primer...

Hello Harry.  See #swig comment

<chimezie> HarryH: in retrospect, there is really nothing about the RDF exracted from HL7 that would be of any use for inference

Below is a somewhat contrived example, but the problem 
primarily is that HL7 RIM is an information model not a knowledge 
representation [1] & [2] So, there is a ceiling on the amount of useful 
inference you can derive from data extracted from HL7 *by itself*.

You could introduce a subclass of foaf:Image for images which depict 
(foaf:depicts [3]) a cpr:medical-sign [4]:


@prefix ex: <,2007-03-20:ImagingSemantics#>.
@prefix cpr: <>.
@prefix foaf: <>.
@prefix owl: <>.
@prefix rdfs: <>.

ex:DiagnosingImage a owl:Class;
                    rdfs:subClassOf foaf:Image,
                    [ a owl:Restriction;
                      owl:onProperty ex:indicates;
                      owl:someValuesFrom cpr:medical-sign ]

ex:indicates a owl:ObjectProperty;
              rdfs:comment "Property relating a foaf:Image to a medical sign it indicates";
              rdfs:subPropertyOf [ owl:inverseOf cpr:interpretant-of ];
              rdfs:domain foaf:Image;
              rdfs:range cpr:medical-sign


So, if you associated the example cpr RDF graph with the 
above OWL fragment and the POMR ontology (there is still no 
consistent precedent on how to do such a thing: associate the ABOX with 
*a* TBOX in OWL), and followed OWL entailment, you could dispatch the following SPARQL query:

PREFIX cpr: <>
PREFIX ex: <,2007-03-20:ImagingSemantics#>
PREFIX skos: <>
SELECT ?sign ?image
   ?image a ex:DiagnosingImage;
          ex:indicates [ skos:prefLabel ?sign ]

This ofcourse, assumes the piece of code responsible for processing this 
bit of SPARQL:

- Is a GRDDL-aware agent (note the GRAPH URL is the source document)
- Can follow its nose (by some yet unknown convention) to find the OWL assertions
- Is capable of performing OWL (DL? Full?) inference *before* dispatching a SPARQL query over an RDF graph

The result:

image  |   sign
_:foo  |  "Chest hyperinflated"

The URI for the image is a BNode in this case.  Not so useful, but 
consider (perhaps) if the medical record instead used a resolvable HTTP 
URI of the actual Chest Xray, or an authoritative identifier from an imaging repository within the 
institution where the Xray was performed.

Hope this suites your needs.


Chimezie Ogbuji
Lead Systems Analyst
Thoracic and Cardiovascular Surgery
Cleveland Clinic Foundation
9500 Euclid Avenue/ W26
Cleveland, Ohio 44195
Office: (216)444-8593

Received on Tuesday, 20 March 2007 19:12:50 UTC