Auditing Redundant Import in Reuse of a Top Level Ontology for the Drug Discovery Investigations Ontology

The use of a top-level ontology, e.g. the Basic Formal Ontology (BFO), as a template for a domain ontology is considered a best practice. This saves design efforts and supports multi-disciplinary research. The Drug Discovery Investigations ontology (DDI) for automated drug discovery investigations followed the best practices and imported BFO. However not all BFO classes were used. Quality assurance is an important process in the development of ontologies. One methodology proven to support quality assurance is based on automatic derivation of abstraction networks (ANs) from the original ontologies. An AN of an ontology is a compact secondary network summarizing the ontology. ANs were shown to support the identification of sets of concepts with higher concentrations of errors than control sets. In this paper, an AN is derived for the DDI, based on object properties. The top node of this AN represents a set of 81 classes without any object properties. Nodes of an AN representing many classes tend to indicate modeling errors. Upon reviewing these 81 classes, we discovered that among them are most of the classes imported from BFO, and that most of these classes are irrelevant for DDI. An algorithm for hiding such irrelevant classes from a specified ontology is described. As many as 18 (56%) of the 32 BFO classes represented by the top node of the AN were hidden from DDI by the algorithm. We conclude that ontologies reusing a top-level ontology should employ this AN-based approach.