Graph Schema

The Resource Description Framework Schema (RDFS) and Web Ontology Language (OWL) are W3C standards for defining semantic schemas. In biomedicine, they create rigorous, machine-interpretable definitions for entities like genes, proteins, and diseases.

• Example: The Gene Ontology (GO) uses OWL to define classes like BiologicalProcess, MolecularFunction, and CellularComponent, with properties like part_of and regulates.
• Implementation: An OWL axiom might state 'Mitosis' SubClassOf 'CellCyclePhase', enabling automated reasoners to infer that any instance of Mitosis is also an instance of CellCyclePhase.
• Use Case: This allows researchers to integrate disparate datasets by mapping them to a common, logically consistent schema, enabling complex queries across genomics and clinical data.

Property graph databases like Neo4j use a label-property model. A schema defines allowed node labels, relationship types, and property constraints.

• Example Schema for a Supply Chain:
  • Node Labels: Supplier, Product, Warehouse, Retailer.
  • Relationship Types: SUPPLIES (Supplier→Product), STORED_IN (Product→Warehouse), SHIPS_TO (Warehouse→Retailer).
  • Property Constraints: A Product node must have a sku (String, unique) and unitCost (Float). A SHIPS_TO relationship must have a shipDate (DateTime).
• Implementation: Neo4j uses Cypher queries for schema introspection and optional APOC procedures to enforce property existence and types. This schema enables efficient pathfinding queries like "Find all suppliers for products out of stock in California."

Apache TinkerPop is a graph computing framework. Its Graph Schema feature provides a runtime, programmable schema for property graphs, defining vertex/edge labels, property keys, and data types.

• Core Elements:
  • PropertyKey: Defines name (age) and data type (Integer).
  • VertexLabel: Defines a vertex type (person) with optional partition keys.
  • EdgeLabel: Defines a relationship (knows) with multiplicity (SIMPLE, MANY2ONE).
• Example: A social network schema would define a person vertex label with property keys name (String) and age (Integer), connected by knows edges with a since (Integer) property.
• Use Case: This schema is enforced by compliant graph databases (e.g., JanusGraph, DataStax Graph) to ensure data consistency across massive-scale graphs, crucial for fraud detection networks.

Shapes Constraint Language (SHACL) is a W3C standard for validating RDF knowledge graphs against a set of conditions. It defines "shapes" that specify constraints on nodes.

• Implementation Pattern:
  • A PersonShape declares that instances of ex:Person must have exactly one ex:ssn property matching a regex pattern, and an ex:worksFor property that points to a node of type ex:Company.
• Example Constraint:

  turtleCopy
  undefined

ex:PersonShape a sh:NodeShape ; sh:targetClass ex:Person ; sh:property [ sh:path ex:ssn ; sh:datatype xsd:string ; sh:pattern "^[0-9]{3}-[0-9]{2}-[0-9]{4}$" ; ] .

codeCopy
- **Use Case**: Enterprises use SHACL to enforce **data quality** and **regulatory compliance** (e.g., GDPR) in their knowledge graphs, ensuring required fields are present and formatted correctly before data is consumed by downstream applications.

SQL Property Graph Queries (SQL:PGQ) is a forthcoming ISO standard extending SQL to define and query property graph views over existing relational tables. This implements a virtual graph schema without migrating data.

• How It Works: A GRAPH schema is defined using CREATE PROPERTY GRAPH statements that map:
  • Vertex Tables: A customers table becomes Customer vertices, with columns like customer_id mapped to a vertex ID and customer_name to a property.
  • Edge Tables: A purchases table becomes PURCHASED edges, linking Customer to Product vertices.
• Example: A financial institution can define a transaction graph over its normalized OLTP schema, instantly enabling graph queries for money laundering detection without ETL. This schema acts as a logical layer for graph analytics on traditional data.

Schema.org is a collaborative, community-driven vocabulary for structuring data on the web. It provides a shared schema of types (e.g., Person, Event, Product) and properties (e.g., name, startDate, price).

• Implementation: Websites implement this schema using JSON-LD, Microdata, or RDFa markup embedded in HTML. For example, a product page includes structured data defining a Product type with name, offers, and brand properties.
• Core Function: It creates a giant, publicly accessible knowledge graph used by search engines, AI assistants, and recommendation systems. The schema ensures consistent interpretation of web content.
• Enterprise Adaptation: Companies extend Schema.org with custom vocabularies (using schema:domainIncludes) to describe proprietary entities (e.g., MedicalDevice) while maintaining interoperability with the global web graph, enhancing SEO and Answer Engine visibility.

An ontology is a formal, explicit specification of a shared conceptualization. It defines a set of concepts, categories, properties, and the relationships between them within a specific domain. While a graph schema provides the structural rules for data storage, an ontology provides the semantic meaning and logical constraints, enabling advanced reasoning and inference.

• Key Components: Classes (types of things), properties (attributes of things), and axioms (logical rules).
• Formal Languages: Typically expressed in standards like Web Ontology Language (OWL) or RDF Schema (RDFS).
• Use Case: In a biomedical knowledge graph, an ontology would define that a 'Gene' encodes a 'Protein', and that this relationship is transitive and asymmetric.

The Property Graph Model is the dominant data structure for modern graph databases. It consists of:

• Nodes (Vertices): Represent entities, which can have labels (e.g., Person, Product) and properties (key-value pairs like name: 'Alice').
• Relationships (Edges): Represent directed connections between nodes, which are always typed (e.g., PURCHASED, WORKS_FOR) and can also have properties (e.g., since: 2020).

A graph schema for a property graph database defines the allowed node labels, relationship types, and the property keys and data types associated with each. This model prioritizes flexibility and efficient traversal.

Resource Description Framework (RDF) is a standard model for data interchange on the web, representing information as triples: Subject-Predicate-Object (e.g., Paris isCapitalOf France).

RDF Schema (RDFS) is a semantic extension that provides a basic vocabulary for describing RDF data. It acts as a primitive schema language, allowing you to define:

• Classes (rdfs:Class) and sub-class hierarchies.
• Properties (rdf:Property) and sub-property hierarchies.
• Domain and range constraints for properties.

While less expressive than OWL, RDFS provides the foundational schema layer for the Semantic Web stack and many knowledge graphs.

Schema.org is a collaborative, community-driven initiative that provides a collection of shared vocabularies (schemas) for structured data markup on the internet. Webmasters use this vocabulary to tag HTML content, enabling search engines and other applications to understand the information on a page.

• Pre-defined Types: Offers a wide range of schemas for entities like Person, Event, Product, and LocalBusiness.
• Relationship to Graph Schemas: The Schema.org vocabulary can be directly used as, or mapped to, the ontology or schema of an enterprise knowledge graph. This allows organizations to align their internal data with a universal standard, facilitating data integration and interoperability with external systems.