Biolink Model

The Biolink Model is an open-source, high-level data model designed to standardize types and relationships within biological knowledge graphs. It provides a consistent framework for representing biological knowledge across diverse databases and formats, specifically targeting translational science.

Key features include:

• Standardized Entities: Covers a wide range of biological categories such as genes, diseases, chemical substances, organisms, genomics, and phenotypes.
• Hierarchical Structure: Uses object-oriented classification through a set of hierarchical, interconnected classes (categories).
• Graph-Oriented Relationships: Defines relationships between entities using predicates (edges).
• Schema Guidance: Provides class and edge attributes and associations that dictate how entities should relate to one another.

To specify the type of a Biolink node in RDF, use the rdf:type predicate to link to a specific node type. Additionally, you MAY use the biolink:category predicate to represent extra categories for that node.

For organizational purposes, rdf:type triples can be partitioned into separate Named Graphs (e.g., placing direct assertions in a main graph and inferred types in an 'inferred' graph).

TRAPI 1.3 supports qualifiers to provide additional context to edges. To implement this support, the following model changes are applied:

• The Edge object includes a qualifiers property.
• A new Qualifier object is defined.
• The QEdge model is extended to include support for qualifier-constraints.

For detailed technical specifications, refer to the TRAPI 1.3 data representation.

Edges represent relationships between nodes. You can attach semantics to edges using three methods:

1. predicate slot: Defines the relationship type. The value must be from the related to hierarchy. Note: Always use the CURIE form with the biolink: prefix (e.g., biolink:interacts_with).
2. category slot: Defines the association type. The value must be from the association hierarchy. Note: Always use the CURIE form with the biolink: prefix (e.g., biolink:GeneToGeneAssociation).
3. type slot: Provides additional semantic typing from external ontologies or taxonomies.

In a graph representation, each entity is a node. To annotate nodes according to the Biolink Model, you can use three different semantic attachment methods:

1. category slot: Use this to assign a primary entity class. The value must be a term from the named thing hierarchy. Note: Always use the CURIE form with the biolink: prefix (e.g., biolink:Protein).
2. type slot: Use this to provide additional semantic typing from any external ontology, controlled vocabulary, or taxonomy (corresponds to rdf:type).
3. subclass_of slot: Use this to define hierarchical relationships using values from external ontologies (corresponds to rdfs:subClassOf).

When multiple qualifiers are applied to a single object, they follow a specific hierarchical order. Each new qualifier layer applies to the concept composed by all preceding qualifiers. This layering prevents ambiguity.

The recommended order for composing an object concept is:

1. object: The core concept (e.g., CDKN2A gene).
2. object_part: A specific part of the core concept (e.g., promoter of the gene).
3. object_form_or_variant: A specific form of that part (e.g., mutant_form of the promoter).
4. object_aspect: A specific aspect of that form (e.g., methylation of the mutant promoter).
5. object_direction: The direction of the aspect (e.g., increased methylation of the mutant promoter).
6. object_context: The context of that direction (e.g., increased methylation of the mutant promoter in the nucleus).

The slot_usage property allows you to provide detailed documentation for how a specific slot should be used within a particular class. This is useful for defining the expected range (the type of value) and providing a description for that slot in the context of the class.

This is particularly helpful for complex classes where a generic slot (like subject) takes on specific semantic meaning.

  gene to gene association:
    is_a: association
    defining_slots:
      - subject
      - object
    slot_usage:
      subject:
        range: gene or gene product
        description: the subject gene in the association.
      object:
        range: gene or gene product
        description: the object gene in the association.

Biolink Association Statements use a two-level semantic structure. When qualifiers are present, you must interpret the statement at two different levels of granularity:

1. Core Triple Reading: This is the high-level relationship. It considers only the subject, predicate, and object slots. This reading must be clear and true (e.g., "A affects B").
2. Full Statement Reading: This provides the detailed context. It combines the core concepts with their respective qualifiers (subject/object qualifiers and statement qualifiers) and uses the qualified_predicate if provided. This reading must also be clear and true (e.g., "A metabolite causes increased B in context C").

The Full Statement Reading should provide additional detail for the core triple claim, rather than representing an unrelated statement.

Predicate slots can define the directionality and symmetry of relationships:

• symmetric: A boolean indicating if the relationship is bidirectional (if A -[r]-> B, then B -[r]-> A). Note: This property is not inherited by descendants; it must be explicitly defined for every symmetric predicate.
• inverse: Specifies the inverse predicate of a given relationship (e.g., if affects is the predicate, affected by is its inverse).

  affects:
    is_a: related to
    inverse: affected by

  interacts with:
    domain: named thing
    range: named thing
    symmetric: true

Knowledge source retrieval provenance describes the chain of Information Resources through which knowledge expressed in a Biolink Edge was retrieved and transformed before reaching its current serialized form (e.g., a TRAPI message).

Key Distinctions:

• Scope: It tracks the mechanical retrieval and transformation of data between web-accessible systems.
• Exclusions: It does not trace knowledge back to specific publications/datasets, nor does it cover the reasoning, inference, or analysis activities used to generate knowledge from evidence.

Example Chain: ARAGORN (ARA) $\rightarrow$ MolePro (KP) $\rightarrow$ ChEMBL (Aggregator) $\rightarrow$ ClinicalTrials.gov (Original Resource)