Skip to content

Standards alignment

How this project's record model relates to the evidence standards biocurators already use. Every term below was checked against its source on 2026-09-25:

Standard Version checked Source
Evidence & Conclusion Ontology (ECO) release 2026-07-10 OLS
Biolink Model 4.4.4 biolink_model.yaml
GA4GH Variant Annotation Specification (VA-Spec) 1.0.1 va-core-source.yaml
W3C PROV-O Recommendation PROV-O

Mapping strength follows SKOS: exact (interchangeable), close (same intent, some difference in scope), related (overlapping idea, not a substitute), none.

Records, statements and evidence

This project GA4GH VA-Spec 1.0.1 Biolink 4.4.4 PROV-O Strength and notes
Statement (subject, predicate, object) Statement with a Proposition (subject, predicate, object) Association (subject, predicate, object) – close. VA separates the proposition from the statement about it; here they are one object.
Statement.scope (explicit tokens) proposition qualifiers, e.g. alleleOriginQualifier, geneContextQualifier association qualifiers – related. Scope is a flat token list compared for equality, not typed qualifiers.
statement_status – – – none.
EvidenceLine EvidenceLine (hasEvidenceItems, directionOfEvidenceProvided) – – close. Biolink does not model evidence lines.
direction: supports / contradicts / neutral supports / disputes / neutral – – exact values; contradicts = disputes.
EvidenceItem an item of EvidenceLine.hasEvidenceItems has evidence (range: information content entity) Entity close to VA; related to Biolink.
EvidenceItem.extracted_text – supporting text – close.
EvidenceItem.extraction_method – agent type – See extraction methods.
SourceArtifact Document (via reportedIn) primary knowledge source / retrieval source Entity (schema class_uri) exact to PROV; close to VA; related to Biolink, which names an information resource rather than a hashed file snapshot.
SourceArtifact version, retrieved_at, sha256, observed_sha256 – – – none. Project extension for frozen-source audit.
Agent (human / software / organization) Agent.agentType (person / software / organization recommended) – Agent (schema class_uri) exact to PROV; close to VA (human = person).
Adjudication (decision, rationale, uses, reviewer, time) Contribution (contributor, activityType, date) agent type: manual_validation_of_automated_agent, when a person accepts automated output – related. No standard field binds a human decision to a specific intended use.
Profiles, use contracts, per-use decisions – – – none. This is what the project adds: the same statement can be admitted for one use and not another.

Extraction methods

Machine-readable in the packaged LinkML schema as meaning: on each ExtractionMethod value (the compiled validation schema and schema_sha256 are unchanged by these annotations).

extraction_method ECO term (release 2026-07-10) Biolink agent type Strength and notes
manual_curation ECO:0000352 evidence used in manual assertion manual_agent close.
deterministic_parser ECO:0000313 imported information used in automatic assertion the upstream source's agent type close to ECO. When the imported record was itself manually asserted (e.g. a ClinVar lab submission), ECO:0000322 imported manually asserted information used in automatic assertion is more specific. Biolink's agent type describes who produced the knowledge, not the importer.
normalized_string_match ECO:0008021 string-matching method evidence used in automatic assertion automated_agent close.
llm_extraction ECO:0008004 machine learning method evidence used in automatic assertion text_mining_agent close. ECO had no LLM-specific term at the release checked. Biolink notes that text-mining agents are prone to misinterpretation and that the source text should be consulted, which is why built-in profiles default to allow_llm_only: false.
(after a human accept adjudication of automated evidence) ECO:0000218 manual assertion, which "could involve human review of computationally generated information" manual_validation_of_automated_agent close, at the level of the assertion rather than the evidence item.

To read the ECO meanings in code:

from linkml_runtime.utils.schemaview import SchemaView
from bioevidence_validator.engine import default_schema_path

view = SchemaView(str(default_schema_path()))
eco = {name: value.meaning for name, value in view.get_enum("ExtractionMethod").permissible_values.items()}
# {'deterministic_parser': 'ECO:0000313', 'manual_curation': 'ECO:0000352', ...}

The ClinVar case in VA-Spec terms

In the ClinVar case, each ClinVar submission (SCV) corresponds to a VA-Spec Statement by its submitter, and the case's statement is related to a VariantPathogenicityProposition (predicate isCausalFor, with an objectCondition). The case is variant-level and does not model the condition, so it is not a VA-Spec pathogenicity statement. Submissions are imported by deterministic_parser; ECO:0000322 describes them more precisely.

Not mapped

  • SEPIO. The schema declares the sepio: prefix from earlier design work, but no SEPIO terms are used. VA-Spec's introduction states that it adopts and builds on the SEPIO model, so alignment is stated against VA-Spec.
  • The word "profile". A SEPIO / VA-Spec profile specializes the data model for a knowledge type. A profile here specializes the admission policy: which evidence each intended use requires. The two are complementary; a VA-Spec pathogenicity statement could be checked against a profile of this project.
  • Biolink knowledge level. Not modeled. A profile can require evidence types that imply a knowledge level, but the engine does not read or set it.
  • Strength and scores. VA strength/score and Biolink confidence values are not used; admission is decided by per-use evidence requirements, not a numeric threshold.