Three W3C standards dominate biomedical knowledge representation: the Resource Description Framework (RDF) for data graphs, the Simple Knowledge Organization System (SKOS) for controlled vocabularies and thesauri, and the Web Ontology Language (OWL) for full logical ontologies. Choosing the right formalism before building saves months of rework later.
RDF: The Foundation Layer
RDF is a general-purpose data model that represents knowledge as subject-predicate-object triples. It is the foundation on which SKOS and OWL are both built. RDF on its own gives you a flexible, linked-data-compatible graph structure — useful for integrating heterogeneous datasets and exposing them via SPARQL endpoints. Its limitation is that it provides no mechanism for defining class hierarchies or asserting constraints; it describes what is, not what must be.
SKOS: The Right Choice for Controlled Vocabularies
SKOS is an RDF vocabulary designed for representing thesauri, classification schemes, and controlled vocabularies. It provides concepts, labels, broader/narrower hierarchies, and related-term links. SNOMED CT, MeSH, and many internal pharmaceutical vocabularies can be represented in SKOS with relatively low effort. If your primary goal is controlled indexing and synonym-aware search — without needing logical inference — SKOS is usually the fastest path to value. Its weakness is that SKOS offers no formal semantics: you cannot define what it means for two concepts to be related, only that they are.
OWL: Full Expressive Power
OWL adds description logic to the RDF foundation, enabling the definition of classes via necessary and sufficient conditions, property restrictions, and cardinality constraints. An OWL reasoner can classify a new individual under the correct concept based solely on its stated properties, detect logical inconsistencies, and infer relationships that were never explicitly asserted. This inferential power comes at a cost: OWL modelling requires expertise in description logics, and reasoning over large biomedical ontologies can be computationally demanding. For most pharmaceutical and clinical knowledge management applications, OWL 2 EL — a tractable profile used by SNOMED CT itself — provides the right balance of expressiveness and performance.
The practical recommendation: start with SKOS if you need a controlled vocabulary today, and design it so that the concept identifiers can become OWL class URIs if you decide to add formal semantics later. The migration from SKOS to OWL is far simpler than rebuilding from scratch.
