We present initial experiments of transforming an expressive OWL DL model into RDF, with the goal of maximally preserving its semantics. Based on a semantic interoperability use case (three different ways of encoding the same medical information), we use OWL DL representation and reasoning results as gold standards and compare them with the results of querying two derived RDF representations enhanced by pre-computing of inferred triples by using the RL profile and approximate custom rules.
Diagnosis Organ Status Caused by ischaemic heart disease
Heart Suspected
Yes No Unknown x
Suspected heart failure caused by ischaemic heart disease
Diagnosis
Heart Failure Status
Suspected Cause
Ischaemic heart disease d3 rdf:type 'Diagnostic statement' and 'is about situation' only 'Heart failure' d3 rdf:type 'Diagnostic statement' and 'has information attribute' some 'Suspected ' d3 rdf:type 'Diagnostic statement' and 'is about situation' only ('Disorder'
and 'is caused by' some 'Ischaemic heart disease')
Anticipating scalability problems of expressive OWL DL T-boxes we approximate
the results obtained by OWL DL reasoning by an alternative approach that uses RDF,
SPARQL and materializations, triggered by axioms from the underlying OWL
ontology. We present two RDF-based representations: (i) the OWL/RDF serialization of
the original OWL DL axioms, and (ii) an RDF representation obtained as a
simplification of the OWL DL one, in which the subject and object of the RDF triple
correspond to ontology classes and the predicate to an OWL DL expression. For the d3
instance, which is of the type 'Diagnostic statement':
<d3> <IS ABOUT SITUATION> <HeartFailure#123>
<d3> <HAS INFORMATION ATTRIBUTE> <Suspected#3826>
<d3> <IS ABOUT SITUATION> <Disorder#9981>
<Disorder#9981> <IS CAUSED BY> <IschaemicHeartDisease#0951>
Each representation is then checked against a set of SPARQL 1.1 queries, by using
Virtuoso 7.0 and the implementation of the OWL 2 RL profile as a set of SPARQL
queries [
and 'is about situation' only ('Organ failure' and 'is included in' some 'Heart') A DL reasoner is able to infer equivalence between 'Heart failure' and the expressions 'Organ failure' and isIncludedIn some Heart. However, in RDF it would be required to create a specific rule that each time that there is a triple ('Organ Failure' 'included in' Heart), then another one including 'Heart Failure' would be also created.
The drawback of the representation (i) is its complexity due to numerous intermediate blank nodes. The main drawback of (ii) is the need of keeping the track of the instances when generating the triples.
Although the OWL 2 RL captures a substantial fragment of OWL 2, it restricts
expressiveness and is not able to capture existentially quantified expressions [
Our work presented here is preliminary and based on a concrete pattern of a semantic interoperability use case that is highly relevant for health care data. Further work will address more use cases and explore other ways of approximating DL reasoning, taking into account recent developments of triple stores.
Acknowledgements. This work was funded by the SemanticHealthNet Network of Excellence within the EU 7th FP, Call:FP7-ICT-2011-7, agreement 288408. http://www.semantichealthnet.eu/