A Computable Framework for Sequence Variant Classification v4 Criteria Specifications within the ClinGen CSpec Registry
ASHG 2026
Neethu Shah, Andrew R. Jackson, Christine Preston, Deborah I. Ritter, Natasha Strande, Pamela A. Robertson, Justyne E. Ross, Danielle Azzariti, Marina DiStefano, Alicia Byrne, Steven. M. Harrison, Sharon E. Plon, Aleksandar Milosavljevic
Abstract
The ClinGen Criteria Specification (CSpec) Editor, Registry, and associated APIs serve as a centralized suite of services for the management and dissemination of ACMG/AMP sequence variant classification criteria specifications defined by Variant Curation Expert Panels (VCEPs). This specification management system (cspec.clinicalgenome.org) has seamlessly enabled the drafting, review, approval, and publication of ACMG/AMP-based criteria specifications and significantly enhanced consistency in usage and transparency of variant classification rationales. To date, the CSpec Registry contains more than 121 approved gene-disease specific criteria specifications, with 85 additional in various stages of the approval process.
To support the forthcoming ACMG/AMP/ClinGen/CAP Sequence Variant Classification v4 (SVCv4) standards, the CSpec services are being enhanced to allow the creation of new SVCv4 criteria specifications and the migration of existing v3 specifications into the updated framework. For this, we have implemented a new computable data model within the CSpec, designed to represent and support the expanded structure, evidence modeling, and quantitative capabilities introduced by the SVCv4 standards. This model represents the SVCv4 standards as a directed acyclic graph, enabling a structured, non-cyclical representation of decision logic. The model adapts concepts from the Workflow Description Language (WDL), an open-source, human-readable framework for defining and executing data processing workflows. It consists of four core entities-Document, Workflow, Task, and Ruleset which are connected through explicitly defined edges that represent entity relations and execution flow into three interdependent layers: a domain layer encoding gene-disease context and evidence relations as structured facts, a rules layer with classification criteria, and an execution layer to apply thresholds and compute classification.
This CSpec SVCv4 data model in alignment with FAIR principles will enable structured, reusable, and computable representation of SVCv4 criteria specifications, supporting consistent and reproducible variant classification across VCEP and wider genomics community.
Related publications