Resource Description Framework (RDF)

RDF breaks all information down into atomic semantic triples, a simple yet powerful data model consisting of a subject, predicate, and object. This structure encodes a single fact about a resource.

• Subject: The resource being described (an IRI or blank node)
• Predicate: The property or relationship (an IRI)
• Object: The value or another resource (an IRI, literal, or blank node)

For example, the statement 'Patient A has a diagnosis of Diabetes' becomes a triple connecting a patient identifier to a disease concept via a 'hasDiagnosis' predicate.

A collection of RDF triples naturally forms a directed, labeled graph. Subjects and objects become nodes, and predicates become the labeled, directed edges connecting them. This graph structure excels at representing highly interconnected, heterogeneous data without requiring a rigid, pre-defined schema.

Unlike relational tables, the graph model treats relationships as first-class citizens, enabling efficient traversal across complex networks of clinical concepts, patient records, and provenance metadata.

RDF mandates the use of Uniform Resource Identifiers (URIs)—a superset of URLs—to uniquely name all resources, predicates, and classes globally. This prevents naming collisions and enables unambiguous data merging across disparate systems.

• A specific patient is identified by a unique URI, not just a local ID
• A medical concept like 'Myocardial Infarction' is referenced by its SNOMED CT URI
• This global scope is the foundation of Linked Data and decentralized knowledge integration

Because every entity and relationship is uniquely identified by a global URI, RDF datasets from multiple sources can be automatically merged without explicit, point-to-point schema mapping. The graph simply expands to incorporate new triples.

This schema-agnostic nature is critical for integrating heterogeneous healthcare data—combining FHIR resources, clinical trial data, and genomic annotations into a single, queryable knowledge graph without a massive upfront ETL transformation.

RDF is an abstract data model, not a file format. It can be serialized into several concrete syntaxes, each optimized for different use cases:

• Turtle (TTL): A compact, human-readable format ideal for authoring and debugging
• JSON-LD: A JSON-based format for embedding linked data in web APIs and pages
• RDF/XML: A legacy XML-based syntax for compatibility with older XML toolchains
• N-Triples: A simple, line-based format for high-volume streaming and processing

RDF itself provides a simple data model, but it is extended by vocabularies like RDFS (RDF Schema) and OWL (Web Ontology Language) to add formal, machine-interpretable semantics. This enables automated reasoning.

A reasoner can infer new, implicit triples from explicit facts and ontological axioms. For instance, if a triple states 'Metformin is an instance of Biguanide' and the ontology defines 'Biguanide is a subclass of Antidiabetic Agent', the reasoner infers 'Metformin is an Antidiabetic Agent' without it being explicitly stated.