Ontology Engineering

Classes define the types of entities that exist in a domain, such as Patient, Medication, or Procedure. They are organized into a taxonomic hierarchy using rdfs:subClassOf relationships, enabling inheritance of properties and facilitating logical reasoning. A well-engineered class hierarchy ensures that a search for a Drug also returns instances of its subclasses like Antibiotic or Beta-Blocker.

Properties define the relationships between individuals or between an individual and a literal value. Object properties link two instances (e.g., hasProcedure linking a Patient to a Surgery), while datatype properties link an instance to a concrete value (e.g., hasAge linking to the integer 42). Defining precise domain and range constraints on properties is critical for maintaining data integrity and enabling automated validation.

Beyond simple taxonomies, axioms are the logical rules that give an ontology its inferential power. Using the Web Ontology Language (OWL), engineers define restrictions like owl:allValuesFrom and owl:someValuesFrom to specify complex class definitions. For example, a HypertensivePatient can be defined as a Patient who has at least one Observation with a code for High Blood Pressure. These axioms allow a reasoner to automatically classify instances.

Instances (or individuals) are the concrete data points that populate the ontology's schema—the actual patients, drugs, and observations. An assertion is a declared fact about an instance, such as Patient123 hasName 'Jane Doe'. The combination of a rich schema (the TBox) and a large volume of instance data (the ABox) transforms an abstract ontology into a queryable, operational knowledge graph.

While OWL axioms are used for inference under the Open World Assumption, validation requires a Closed World Assumption. The Shapes Constraint Language (SHACL) is a W3C standard used to define 'shapes' that instance data must conform to. A SHACL shape can enforce cardinality (e.g., a Patient must have exactly one date of birth), data types, and value ranges, ensuring the knowledge graph's data quality before it is used in a clinical decision support system.

A single ontology rarely exists in isolation. Ontology alignment is the process of creating logical mappings between equivalent or related concepts in different ontologies (e.g., mapping a class in a local hospital ontology to a SNOMED CT concept). These mappings, expressed using predicates like owl:equivalentClass or skos:exactMatch, are the engine of semantic interoperability, allowing a query written against one standard to be automatically rewritten for another.

The process of determining semantic correspondences between concepts from disparate medical terminologies to achieve interoperability. This involves mapping equivalent classes across SNOMED CT, ICD-10-CM, LOINC, and RxNorm. Automated alignment algorithms use lexical matching and structural graph analysis to generate candidate mappings, which are then curated by clinical informaticists to create a unified semantic layer over heterogeneous EHR systems.

The task of mapping diverse textual mentions of a clinical concept to a single, canonical identifier in a standardized vocabulary. For example, the strings 'high blood pressure', 'HTN', and 'elevated BP' must all normalize to the same SNOMED CT code. This resolves synonymy and ambiguity, enabling consistent data aggregation for cohort analysis and population health studies. It is a prerequisite for accurate ontology population.

A software component that infers logical consequences from a set of asserted facts and ontological axioms. Using deductive processes, a reasoner can derive new implicit knowledge, such as: if 'Myocardial Infarction' is a subclass of 'Ischemic Heart Disease', and a patient is diagnosed with an MI, the reasoner infers they also have a cardiovascular disorder. This is essential for knowledge graph completion and consistency checking.

The NLP task of identifying a textual mention of an entity in unstructured text and grounding it to its corresponding unique entry in a knowledge base. In a clinical note mentioning 'metformin', entity linking resolves it to the specific RxNorm ID for the drug, distinguishing it from a misspelling or a brand name. This process bridges unstructured narrative text with the structured, formal ontology.