find it difficult to repair errors in an ontology and may even start to doubt the correctness of inferences. Such an explanation typically involves computing a (hopefully small) subset of the ontology that still entails the inference in question and, if necessary, presenting the user with a chain of reasoning steps. ¹² The explanation in Figure 3 (produced by the Protégé 4 ontology-development tool) describes the steps that lead to the inference mentioned earlier with respect to the inconsistency of OceanCrustLayer.

ontology applications

The availability of tools and reasoning systems has contributed to the increasingly widespread use of OWL, which has become the de facto standard for ontology development in fields as diverse as biology, ¹⁹ medicine, ¹⁸ geography, ⁸ geology (the Semantic Web for Earth and Environmental Terminology project, sweet.jpl.nasa.gov/), agriculture, ²⁰ and defense. ¹⁵ Applications of OWL are particularly prevalent in the life sciences where OWL is used by developers of several large biomedical ontologies, including SNOMED, GO, and BioPAX, mentioned earlier, as well as the Foundational Model of Anatomy ( sig.biostr. washington.edu/projects/fm/) and the U.S. National Cancer Institute thesaurus ( www.cancer.gov/cancertopics/ter-minologyresourceshow).

The ontologies used in these applications might have been developed specifically for the purpose or without any particular application in mind. Many ontologies are the result of collaborative efforts within a given community

aimed at facilitating (Web-based) information sharing and exchange; some commercially developed ontologies are also subject to a license fee. Many OWL ontologies are available on the Web, identified by a URI and should, in principle, be available at that location. There are also several well-known ontology libraries and even ontology search engines (such as SWOOGLE, swoogle.umbc.edu/) that are useful for locating ontologies. In practice, however, applications are invariably built around a predetermined ontology or set of ontologies that are well understood and known to provide suitable coverage of the relevant domains.

The importance of reasoning support in ontology applications was highlighted in a paper describing a project in which the Medical Entities Dictionary (MED), a large ontology ( 100,210 classes and 261 properties) used at the Columbia Presbyterian Medical Center in New York, was converted to OWL and checked using an OWL reasoner. ¹³ As reported in the paper, this check revealed “systematic modeling errors” and a significant number of missed subClassOf relationships that, if not corrected, “could have cost the hospital many missing results in various decision-sup-port and infection-control systems that routinely use MED to screen patients.”

In another application, an extended version of the SNOMED ontology was checked using an OWL reasoner that found a number of missing subClassOf relationships. This ontology is being used by the U.K. National Health Service (NHS) to provide “a single and comprehensive system of terms, cen-

trally maintained and updated for use in all NHS organizations and in research” and as a key component of its $6.2 billion “Connecting for Health” IT program ( www.connectingforhealth. nhs.ukhow). An important feature of the system is that it can be extended to provide more detailed coverage if needed by specialized applications; for example, a specialist allergy clinic may need to distinguish allergies caused by different kinds of nut so may need to add new terms to the ontology (such as AlmondAllergy):

Class: AlmondAllergy equivalentTo: Allergy and causedBy some Almond

Using a reasoner to insert this new term into the ontology ensures it is recognized as a subClassOf NutAllergy, something that is clearly of crucial importance for ensuring that patients with an AlmondAllergy are correctly identified in the national records system as patients with a NutAllergy.

Ontologies are also widely used to facilitate the sharing and integration of information. The Neurocommons project ( sciencecommons.org/proj-ects/data/) aims to provide a platform for, for example, sharing and integrating knowledge in the neuroscience domain; a key component is an ontology of annotations to be used to integrate available knowledge on the Web, including major neuroscience databases. Similarly, the Open Biomedical Ontologies Foundry ( www.obofoundry.org) is a library of ontologies designed to facilitate international information sharing and integration in the biomedical domain. In information-integration ap-