Prevalence and Effects of Class Hierarchy Precompilation in Biomedical Ontologies

It is sometimes claimed that adding inferred axioms, e.g. the inferred class hierarchy (ICH), to an ontology can improve reasoning performance or an ontology’s usability in practice. While such beliefs may have an effect on how ontologies are published, there is no conclusive empirical evidence to support them. To develop an understanding of the impact of this practice, both for ontology curators as well as tools, we survey to what extent published ontologies in BioPortal already contain their ICH and most specific class assertions (MSCA). Furthermore, we investigate how added inferred axioms from these sets can affect the performance of standard reasoning tasks such as classification and realisation. We find that axioms from the ICH and MSCA are highly prevalent in published biomedical ontologies. Our reasoning evaluation indicates that added inferred axioms are likely to be inconsequential for reasoning performance. However, we observe instances of both positive as well as negative effects that seem to depend on the used reasoner for a given ontology. These results suggest that the practice of adding inferred axioms during the release process of ontologies should be subject to a task-specific analysis that determines whether desired effects are obtained.