When building a knowledge infrastructure that must integrate ontologies from multiple biomedical domains — genetics, pharmacology, anatomy, clinical trials, adverse events — a fundamental question arises: how do you ensure that the concept of process in a genomics ontology means the same thing as the concept of process in a clinical procedures ontology? The answer lies in upper ontologies.
What Upper Ontologies Provide
An upper ontology (also called a foundational ontology) defines the most general categories that apply across all domains: entities and processes, continuants and occurrents, material and immaterial things, dispositions and functions, roles and qualities. By anchoring domain ontologies to a shared set of foundational categories, upper ontologies make cross-domain reasoning coherent. A drug substance in a pharmacology ontology and a chemical entity in a chemistry ontology can both be grounded as material entities with specific dispositions — and a reasoner can then draw inferences that cross the original domain boundary.
BFO and the OBO Foundry
The Basic Formal Ontology (BFO) is the most widely adopted upper ontology in biomedicine. It underpins the entire Open Biomedical Ontologies (OBO) Foundry collection, which includes the Gene Ontology, the Disease Ontology, the Ontology for Biomedical Investigations, ChEBI for chemical entities, and dozens of others. By committing to BFO as a shared upper layer, OBO Foundry ontologies can be merged and reasoned over together — a significant practical advantage for projects that need to integrate multiple biomedical knowledge sources.
Practical Considerations for Integration Projects
For most pharmaceutical and clinical knowledge management projects, the direct commitment to BFO is less important than the structural discipline it enforces: clearly distinguishing between entities and processes, between roles played by things and intrinsic properties of things, and between what is universally true and what is merely contingently true. These distinctions prevent the most common modelling errors — conflating a disease with its manifestation, confusing a drug role with a drug substance, or representing a clinical finding as if it were a biological entity. Whether or not your project explicitly imports BFO, modelling with these principles in mind produces ontologies that integrate far more easily with external knowledge sources.
