Graph Database

The dominant data model underpinning native graph databases like Neo4j. Unlike RDF's triple structure, the property graph model represents data as nodes (entities) and relationships (edges), both of which can hold an arbitrary set of key-value pair properties.

• Nodes represent entities (e.g., Patient, Medication, Physician)
• Relationships are typed and directional (e.g., :PRESCRIBED, :HAS_ALLERGY)
• Properties store metadata directly on the element (e.g., {name: 'Lisinopril', dosage: '10mg'})

This model provides an intuitive, whiteboard-friendly structure for representing complex, highly connected clinical data without the need for rigid schemas or expensive JOIN operations.

A declarative, pattern-matching graph query language originally developed by Neo4j and now an open standard (openCypher). Cypher uses ASCII-art syntax to express graph traversals visually.

• MATCH (p:Patient)-[:HAS_DIAGNOSIS]->(d:Diagnosis) RETURN p, d
• Designed for expressive, efficient pathfinding and subgraph extraction
• Analogous to SQL for relational databases, but optimized for graph pattern matching

Cypher allows developers to write complex multi-hop queries that would require dozens of SQL JOINs in a single, readable statement, making it ideal for clinical cohort identification and drug interaction analysis.

A W3C standard data model for representing metadata and knowledge as directed, labeled graphs using semantic triples. Each triple consists of a subject-predicate-object structure that encodes a single atomic fact.

• Subject: The resource being described (e.g., a specific drug)
• Predicate: The property or relationship (e.g., hasGenericName)
• Object: The value or related entity (e.g., 'Metformin')

RDF is the foundational model for the Semantic Web and is serialized in formats like Turtle, RDF/XML, and JSON-LD. It excels at data interchange and integration across heterogeneous healthcare systems.

The standard recursive-acronym query language for retrieving and manipulating data stored in RDF triplestores. SPARQL is to graph databases what SQL is to relational databases, but it operates on graph pattern matching rather than table joins.

• SELECT ?drug WHERE { ?drug rdf:type :Medication . ?drug :treats :Hypertension }
• Supports federated queries across distributed endpoints
• Includes CONSTRUCT, ASK, and DESCRIBE query forms

SPARQL is essential for querying linked open data and clinical knowledge graphs built on RDF, enabling semantic reasoning over ontologies like SNOMED CT and the Gene Ontology.

A class of deep learning architectures designed to operate directly on graph-structured data. GNNs learn low-dimensional vector representations (embeddings) of nodes by recursively aggregating feature information from neighboring nodes through a message-passing scheme.

• Graph Convolutional Networks (GCNs) generalize convolution to irregular graph domains
• Graph Attention Networks (GATs) use self-attention to weigh neighbor importance
• Applications include drug-drug interaction prediction, patient outcome forecasting, and protein function classification

GNNs transform the explicit symbolic knowledge in a graph database into dense numerical representations suitable for predictive machine learning pipelines.

A purpose-built database management system optimized for the storage, indexing, and retrieval of RDF data in the form of semantic triples. Triplestores are the native persistence layer for RDF knowledge graphs.

• Examples: GraphDB, Amazon Neptune, Stardog, Virtuoso
• Optimized for SPARQL query execution with specialized indexing structures
• Support inference and reasoning over ontological axioms

Unlike general-purpose graph databases that use the property graph model, triplestores are designed specifically for the semantic web stack and are commonly used in clinical terminology servers and biomedical knowledge bases.