Querying Large and Expressive Biomedical Ontologies

Real-world biomedical ontologies like the Foundational Model of Anatomy (FMA) are often of large size and can fall in the category of OWL 2 Full. Meta-modeling is commonly adopted in the FMA with most of its classes and roles also specified as individuals, leading to undecidability in reasoning. The state-of-the-art reasoning and querying systems like Pellet often ignore the essence of such meta-modeling. In this paper we present a novel, effective approach to querying large and expressive medical ontologies. More precisely, a sub-language of OWL 2 Full is defined, denoted as Hi(SRIQ), to extend the DL language SRIQ with meta-modeling so that one name can be used as class, role and individual at the same time. We give an algorithm to construct multiple SRIQ knowledge bases from a Hi(SRIQ) ontology so that its reasoning and querying tasks can be reduced to be performed in these SRIQs in an independent way. Moreover, a meta-query language is designed to allow for variables present at class/role positions in query. Answering meta-query in constructed SRIQs is transformed into evaluating a group of conjunctive queries over single SRIQ through query partitioning and heuristic materialization of class/role variables in query. As a result, the Hi(SRIQ) version of the FMA yields eight SRIQs and an empirical study shows that our approach outperforms Pellet in query answering. Our work demonstrates that through divide-and-conquer large and expressive ontologies can be queried in both sound and practical way.