A reference for the Classification Model of the forthcoming ACMG/AMP/CAP/ClinGen Sequence Variant Classification v4 (SVCv4) Standards, expressed as a GA4GH GKS VA-Spec community profile.
Status: early development. The SVCv4 Standards (jointly sponsored by ACMG, AMP, CAP, and ClinGen) are being drafted and piloted by the genomic-medicine community, with publication targeted for October 2026 in Genetics in Medicine (GIM). This repository will publish the engineering-facing data-model documentation that accompanies the guidelines' release.
SVCv4 is realised in software through two complementary models:
| Model | What it represents | Where it is published |
|---|---|---|
| Classification Model | The shape of a Variant Pathogenicity Classification — Statements, Propositions, Evidence Lines, and the Evidence Items that support them. What a classification is. | This repository (as a VA-Spec community profile). |
| Method Model | The methods and rules that evaluate the evidence items and data points carried by the Classification Model and produce workflow-specific scores for the version of SVCv4 being applied (baseline or specialised) to a given (VBC, MDE) curation. Encompasses both the baseline SVCv4 methods/rules and the forthcoming domain-specific specialised versions. | ClinGen Criteria Specification (CSpec) — outside this repo and outside GA4GH GKS VA-Spec. |
This repository covers only the Classification Model. The Method Model — the methods and rules that evaluate the evidence captured during a curation and produce workflow-specific scores for the version of SVCv4 being applied (baseline or VCEP-specialised), along with the gene-disease-MOI scoping that determines which specialised version applies — is the concern of ClinGen CSpec, where SVCv4 specification documents and APIs will be published. Both the baseline SVCv4 methods/rules and the forthcoming VCEP specialisations live in CSpec, not in this repository.
The two models meet through method codes and evidence codes that this repo's Evidence Lines carry as references. The Classification Model names methods and evidence codes; the Method Model in CSpec defines what they do.
The data flow between them runs both directions: the Classification Model carries the Evidence Items and data points that a curator captures for a (VBC, MDE), and those are provided to CSpec's methods/rules for the chosen specification version. CSpec evaluates and produces a workflow-specific score; the result lands back in the Classification Model as an Evidence Line that records the method/rule code invoked, the evidence used, and the score/strength produced.
All structural choices in this repository are grounded in the Global Alliance for Genomics and Health (GA4GH) Genomic Knowledge Standards (GKS) workstream. ClinGen is a GA4GH driver project.
The primary GKS dependency for this repo is the Variant Annotation Specification (VA-Spec) (released v1.0). VA-Spec provides a baseline set of classes — Statement, Proposition, EvidenceLine, InformationEntity, EvidenceData — for representing evidence-based scientific assertions about genomic variants. Community profiles layer additional constraints on top of those baseline classes to enforce alignment with the terminology conventions of a specific community guideline.
This repository authors the VA-Spec SVCv4 Community Profile: the SVCv4-specific shape on top of the VA-Spec baseline. That profile is what enables consistent representation, comparison, and processing of SVCv4 classifications with standardised tooling across organisations.
Supporting GKS schemas:
- VRS (Variation Representation Specification) — v2.0. Used to describe the VBC (the variant under consideration).
- Cat-VRS (Categorical Variation Representation) — v1.0. Used where the variant is described categorically.
- gks-core — common identifier, CURIE, and provenance constructs shared across GKS.
ClinGen's Evidence Repository (ERepo) and its JSON-LD SEPIO implementation informed VA-Spec; this project further informs the SVCv4 community profile being layered on top of VA-Spec.
The SVCv4 Standard authored by the SVCv4 working group is intentionally a baseline. It is designed so that disease-domain experts — including ClinGen Variant Curation Expert Panels (VCEPs) — can author specialised versions on top of it:
- Customised methods and scoring rules within workflows.
- Gene/disease scope (gene-disease-MOI) for which a specialisation is applicable.
These specialisations do not exist yet — the baseline itself is still being finalised — but the design accounts for them. Both the baseline SVCv4 methods/rules and the forthcoming VCEP specialisations live in ClinGen CSpec (the Method Model side), not in this repository.
In practice, the community will use the baseline version to define their VBC-MDE classification processes from the SVCv4 publication onward, and apply specialised versions only as they become available for specific gene/disease scopes. The baseline is therefore the operative SVCv4 version until — and wherever — no specialised version applies; specialisations are progressive enhancements layered over it.
So there are two distinct layerings at play; do not confuse them:
- Interoperability layering — VA-Spec baseline classes → SVCv4 community profile (authored here).
- Domain layering — SVCv4 Standard baseline → VCEP specialisations (authored in CSpec).
The Classification Model is expressed using VA-Spec's canonical entity names:
- A Statement carries a Proposition and a final score, plus attributes for strength-direction, score-classification, method (reference), contribution, and a collection of Evidence Lines.
- A Proposition is structured as SPOQ: Subject (the VBC — Variant Being Considered, expressed as a VRS Variation), Predicate (the asserted relationship), Object (the MDE — Mendelian Disease Entity), and Qualifier(s).
- A Variant Pathogenicity Classification is the categorical classification (a position on the Benign ↔ Pathogenic spectrum) produced by the Statement's score.
- An Evidence Line captures the result of invoking a CSpec method/rule on a curator's evidence: it records the method/rule code that was invoked, the Evidence Items used as inputs, the score (and optional strength) produced, and supporting metadata (contribution, optional strength-direction). It is the artifact through which a specific CSpec evaluation enters the Classification Model.
- An Evidence Item / Evidence Data is a single structured datum captured by the curator. Evidence Items are inputs provided to CSpec methods/rules for the chosen specification version; the resulting Evidence Line records which Items were used and the score produced.
Any process, rule, or method that produces a score maps to a VA-Spec Evidence Line. A CSpec workflow takes the Evidence Items captured in the Classification Model, applies its rules under the chosen specification version, and produces a score. The result lands in the Classification Model as an Evidence Line that records the method/rule code, the evidence used, and the score (and optional strength) produced. The workflow itself — its rules, scoring logic, and version selection — lives in CSpec.
Curators and scientists engage with SVCv4 through a Summary Table that organises evidence lines top-down as:
- Evidence Categories (e.g., Human Observational Data, Variant Impact Data)
- Evidence Concepts
- Evidence Codes — each Evidence Code is the jumping-off point for a workflow (in CSpec). For a given (VBC, MDE) curation, the curator captures Evidence Items under an Evidence Code; those items are provided to the workflow, which evaluates them under the chosen specification version and produces an Evidence Line carrying the resulting score.
These terms are canonical SVCv4 vocabulary and appear throughout the guidelines. This repository's data model represents the classifications and evidence lines produced via that summary table; the CSpec system will represent the workflows themselves.
A single integrated reference site that combines:
- Pydantic data models — the Python source of truth for the SVCv4 Classification Model as a VA-Spec community profile.
- JSON Schemas — emitted automatically from the Pydantic models so any language or toolchain can validate and exchange SVCv4 classifications.
- Worked examples — concrete classifications of representative (VBC, MDE) pairs that downstream implementers can test against.
- Narrative documentation — concept guides explaining the model, the VA-Spec layering, the Classification ↔ Method-Model boundary, and the SVCv4 baseline + specialisation design.
The site will be published to the public web alongside (or before) the SVCv4 guideline release in October 2026, and tracks the draft GA4GH VA Community Profile for SVCv4 (JSON Schema and documentation, expected after GA4GH review in 2026).
This project prioritises the Interoperability and Reusability dimensions of the FAIR data principles:
- Interoperability — VA-Spec-grounded schemas, machine-readable JSON Schemas, stable identifiers.
- Reusability — open licence, versioned models, worked examples, clear documentation aimed at implementers.
Software engineers, bioinformaticians, and platform teams in:
- Clinical diagnostic laboratories implementing SVCv4 in production pipelines.
- Translational genomics research groups and consortia.
- Knowledge-base curators exchanging variant classifications.
- Tool builders producing user interfaces, decision support, or automated curation systems.
This is engineering reference documentation. For clinical guidance on applying SVCv4, refer to the forthcoming SVCv4 guideline publication itself. For the method definitions and workflows that compute SVCv4 scores, refer to ClinGen CSpec.
Now (May 2026):
- This README.
- The scaffold design document at
docs/plans/2026-05-19-initial-scaffold.md— repository structure, tooling choices, VA-Spec interop strategy, and verification approach.
Next:
- Python package skeleton (
src/svcv4_model/) with VA-Spec-aligned Pydantic classes (Statement,Proposition,EvidenceLine,EvidenceItem,VariantPathogenicityClassification,VBC,MDE). - JSON Schema export pipeline.
- MkDocs Material site with auto-generated model reference.
- First worked example validated in CI.
Later (tracking the SVCv4 publication trajectory):
- Fill in real SVCv4 Evidence Category / Concept / Code references as the Summary Table stabilises.
- Publish drafts of the VA-Spec SVCv4 community profile in step with GA4GH review.
- Synchronise with the SVCv4 publication (~October 2026, Genetics in Medicine).
This work is produced by the SVCv4 Standards data modeling team, a task force offshoot of the ClinGen Data Platform Working Group, meeting weekly since July 2024.
Key contributors:
- Alicia Byrne, PhD — Broad Institute
- Larry Babb — Broad Institute
- Christine Preston, PhD
- Neethu Shah — ClinGen
Contributors:
- Matt Wright, PhD
- Gloria Cheung
- Bryan Wulf
- Mark Mandell
- Hannah Dziadzio
- Liam Mulhall
The VA-Spec SVCv4 community profile work is conducted in collaboration with the GA4GH GKS VA-Spec authors (Matt Brush, Alex Wagner, Larry Babb).
| Term | Meaning |
|---|---|
| SVCv4 | Sequence Variant Classification v4 — the ACMG/AMP/CAP/ClinGen joint Technical Standard, succeeding the 2015 Richards et al. guidelines (SVCv3). Points-based; replaces v3's strength-categories + combining-rules approach. |
| VBC | Variant Being Considered — the specific germline variant under evaluation; expressed as a VRS Variation. |
| MDE | Mendelian Disease Entity — the disease the VBC is being assessed against. |
| Classification Model | The shape of a classification (Statements, Propositions, Evidence Lines, Evidence Items). Authored by this repo as a VA-Spec community profile. |
| Method Model | The definitions of methods, workflows, criteria, scoring rules. Authored in ClinGen CSpec; not in this repo, not in VA-Spec. |
| Statement / Proposition / Evidence Line / Evidence Item | VA-Spec core entities; see the conceptual-model section above. |
| SPOQ | Subject / Predicate / Object / Qualifier — the structure of a VA-Spec Proposition. |
| Evidence Category / Concept / Code | Canonical SVCv4 vocabulary for the user-facing Summary Table; the Evidence Code is the jumping-off point for a workflow. |
| Workflow | A prescriptive procedure (defined and applied in CSpec) for evaluating the evidence captured under an Evidence Code and producing a score for the version of SVCv4 being applied. Each workflow's result surfaces in the Classification Model as an Evidence Line. |
| CSpec | ClinGen Criteria Specification system — registry and APIs where SVCv4 methods, workflows, and VCEP specialisations are published. |
| VCEP | Variant Curation Expert Panel (ClinGen). Authors specialised SVCv4 method definitions in CSpec. |
| Community Profile (VA-Spec) | A layer of additional constraints on top of VA-Spec's baseline classes to enforce a community's terminology and conventions. |
| VA-Spec / VRS / Cat-VRS / gks-core | GA4GH GKS sub-schemas. VA-Spec is this repo's primary dependency. |
| CSpec / ERepo / SEPIO | ClinGen Criteria Specification; ClinGen Evidence Repository; Scientific Evidence and Provenance Information Ontology. SEPIO informed VA-Spec. |
| GA4GH | Global Alliance for Genomics and Health. |
| GKS | Genomic Knowledge Standards — a GA4GH workstream. |
| GIM | Genetics in Medicine, the journal where SVCv4 will publish. |
| FAIR | Findable, Accessible, Interoperable, Reusable. |
Contribution guidelines will land alongside the first scaffolded code drop. Until then, please open an issue to discuss any proposed direction.
Code is licensed under the terms in LICENSE. Documentation content is intended to be published under a permissive open licence; final licence terms for narrative content will be confirmed before the first published release of the site.