Investigating One Million XRefs in Thirthy Ontologies from the OBO World

Abstract : The more ontologies are produced, the more need to identify mappings between them becomes important. Several practices and tools have been developed to support ontology alignment, but still, it remains a challenge. In the OBO world, ontology developers use cross reference annotations (formally using the oboInOwl:hasDbXRef property) to link a term to an external entity, including another term in another ontology (i.e., a mapping). These XRefs contains information of high value, because they were manually declared/verified by domain experts when the ontology was built. However, extracting and exploiting them remains a challenge for users due to their heterogeneous and chaotic descriptions. In this paper , we report on analysis of approximately 1 million XRefs in 30 ontologies from the OBO world. Our results show that 10.7% of these XRefs are ontology map-pings, but confirm that semantically-ambiguous usage of the cross-reference property which make it impractical or even impossible to reuse. We describe and quantify several issues identified (e.g., different kind of XRefs, miscellaneous formatting , ambiguous targets), and discuss some way to mitigate them especially by using more relevant specific properties taken from standard semantic Web metadata vocabularies.