ga4gh.org

GKS: the fellowship of the VRS and VA data, from alpha to beta

Larry Babb & Matt Brush

ga4gh.org

Agenda (draft)

• Comparison of top level elements of VA Framework against GK-Pilot 2.0, gnomAD and maveDB (v1.3 applications) schemas with a focus on closing the gaps and defining an MVP scope for the VA framework to establish a v1.0alpha baseline.

ga4gh.org

VRS population frequency

ga4gh.org

The GA4GH Variant

Annotation Specification

GA4GH Connect

Part I: Overview and Updates

​

September.19.2023

VA Leads: Matthew Brush, Javi Lopez

VR Leads: Larry Babb, Alex Wagner

GKS Chairs: Bob Freimuth, Andy Yates

GKS Secretariat: Reggan Thomas

ga4gh.org

• The Variant Annotation Specification (VA-Spec) will provide standard models for exchanging knowledge about genetic variation, along with supporting evidence and provenance information.
• The VA-Spec will also provide a modeling framework to enable implementation-driven development of custom schema from which the VA standard models will evolve.
• The VA-Spec will support different levels of adoption and use - from direct contribution to model development and testing, to out-of-the-box adoption of final VA standards.
• VA products are developed in close collaboration with Driver Projects (DPs) providing real-world use cases, domain expertise, implementation models, and testing environments.

VA Spec Overview

VA Driver Projects

ga4gh.org

A landscape analysis identified many different types of annotation statements, and provided modeling requirements specific to each

​

​

• Biological and Clinical Annotations hold knowledge-level statements about a variant independent of any case-level context → primary remit of VA-Spec
• Case-Level Annotations concern observation of a variant in a particular patient → remit of other standards (e.g. Phenopackets, HL7-Clinical Genomics IM, FHIR)

I. Biological VA Statement Types Molecular Consequence

Functional Impact

Population Frequency

Relative Location

Evolutionary Conservation

Variant-Condition Co-Segregation

Variant Immunogenicity

II. Clinical VA Statement Types

Pathogenicity Classification

Therapeutic Response Classification

Diagnostic Classification

Prognostic Classification

Phenotypic Feature Association

Condition Risk

Relative Population Frequency

III. Case-Level Statements

Variant Case Observation

Variant Origin

Variant Clonality

Causal Patient Variant

Variation Statement Types

ga4gh.org

A Statement-Centric Model

• In the VA data, each knowledge assertion is captured in a self-contained Statement object
• The semantics of what is asserted to be true is explicitly structured in terms of a subject, predicate, object, and qualifier(s) (the statement’s ‘Proposition’)

​

“Germline HRAS:c.173C>T is pathogenic for Costello Syndrome”

• The types and granularity of E/P information can vary dramatically across statement types and sources

• Organization of variant knowledge into discrete Statement objects allows clear and precise tracking of the evidence and provenance (E/P) that supports each

Classes and relationships comprising the backbone of the VA Core Information Model

ga4gh.org

Molecular Consequence Statement Profile

Therapeutic Efficacy Statement Profile

Pathogenicity Statement Profile

Therapeutic Response Statement Profile

Profiling Tasks

• Select relevant subset of classes and attributes from the full Core IM
• Create domain-specific Specializations of core classes (e.g. Therapeutic Efficacy Statement)
• Define/reuse models for relevant Domain Entities (e.g. Variation, Treatment, Disease)
• Build Value Sets holding relevant domain concepts
• Apply application-specific data type and value constraints
• Define custom Extensions for new/unsupported content

Statement Profiling

Models for a particular Statement type are defined through a ‘Profiling’ process that specializes the Core IM with domain-specific features and constraints

The VA Core Information Model (Core IM)

A SEPIO-based Core IM provides ‘building blocks’ for profile development

profiling

profiling

profiling

ga4gh.org

The Proposition Assessment Paradigm (SPOQ-DS)

Statements assess the direction and strength of the Proposition at their core

Variant Pathogenicity Example (ClinVar):

Proposition:

• subject: NM_004700.4:c.803CCT[1]
• predicte: isCausalFor
• object: Hearing Loss
• qualifier: [high | low | risk allele ]

​

Assessment:

• direction: [supported | refuted | uncertain]
• strength: [ likely | definitive … ]
• score: float
• conclusion (classification): likely pathogenic

​

A plain language reading:

“NM_004700.4:c.803CCT[1] is causal for Hearing Loss” is assessed as likely supported

ga4gh.org

The Proposition Assessment Paradigm (SPOQ-DS)

The assessed Proposition can collapse into a Statement for a flatter model

Variant Pathogenicity Example (ClinVar):

Statement (collapsed with Proposition):

• subject: NM_004700.4:c.803CCT[1]
• predicte: isCausalFor
• object: Hearing Loss
• qualifier: [high | low | risk allele ]
• direction: [supported | refuted | uncertain]
• strength: [ likely | definitive … ]
• score: float
• conclusion (classification): likely pathogenic

​

A plain language reading:

“NM_004700.4:c.803CCT[1] is causal for Hearing Loss” is assessed as likely supported

ga4gh.org

The Proposition Assessment Paradigm: ClinVar

ga4gh.org

VA-Spec Components (models and code)

1. A Core Information Model (IM). A domain-agnostic information model for structuring knowledge statements and their supporting evidence and provenance.
2. Standard Profiles: Statement-specific ‘profiles’ of the Core IM, providing standards for exchange of specific types of variant knowledge, built on the same underlying model
3. Reference Implementation(s). A library of software and services that demonstrate the creation, validation, and exchange of compliant data using Standard Profiles

Implementation Support (guidelines, processes, and infrastructure)

1. Profiling Methodology: guidance, examples, and tools for executing the profiling process to produce implementation models.
2. Implementation Sandbox: a community testbed for defining, adapting, and sharing implementation profiles
3. Standards Development Process: formal processes through which GA4GH standards evolve from implementation profiles through community collaboration and consensus.

A Framework to Support Distributed Development

ga4gh.org

A Simple Pathogenicity Statement Profile

• Defined as part of a “GK-Pilot project by a joint ClinGen - VICC initiative, to support re-representation, aggregation, and exchange of ClinVar pathogenicity classification data
• Diagram shows a view of a draft minimal model
• Includes a subset of classes and attributes from the Core IM, specialized with domain/application specific content (green):
• Defines/imports 'Domain Entities' (e.g. Variation, Condition)
• Collapses classes with 1:1 dependencies to flatten the model (e.g. Proposition into Statement)
• Defines specializations of Core classes (e.g. Statement --> VariantPathogenicity)
• Specialties/gives new names to attributes (e.g. VariantPath.variation)
• Adds domain/application-specific attribute bindings to classes and code sets (e.g. VariantPath.variation, Contribution.activity)
• Modifies attribute cardinalities (e.g. Contribution.activity: Coding)
• Changes attribute data types (e.g. Activity.date: dateTime -> string)
• Adds new attributes (e.g.V.varPAth.penetrance)

​

See here for more detail about how this Profile was derived from the Core IM

ga4gh.org

Deriving the Pathogenicity Statement Profile

See here for Lucidchart source.

Profiling steps taken to derive a GK-Pilot Pathogenicity Profile from the Core Information Model

ga4gh.org

VA-Spec Development Approach

Implementation-Driven Development

• VA provides the Core IM and a Profiling Framework
• Driver projects define ‘Implementation Profiles’ (IP) for specific statement types
• VA oversees coalescence of implementation profiles for a given statement type into a final ‘Standard Profile’ (SP)

Benefits

• Alleviates development bottleneck: leverage community resources, expertise
• Appeal to different user bases: adopters can use standards out of the box (basic user), or define custom profiles for local applications (power user)
• No surprises at implementation time: model development happens in context of use
• Proven models: standard demonstrated value in working applications

Challenges

• Complexity of infrastructure needed to support community-driven development
• Dependence on implementations introduces a potential new bottleneck

An Implementation-Driven Approach to Profile Definition and Standards Development

ga4gh.org

Implementation-Driven Development

(link to source)

Coordinated Evolution of VA Implementation Profiles and Standards

ga4gh.org

Contexts of Profile Development and Use

(link to source)

'Basic Users' may choose to simply adopt the Standard Profile for purposes of global data exchange (but not apply VA locally/internally).

'Power Users' can define implementation profiles of the Core IM that are suited to their local implementation context (e.g. IP1, IP2).

• These local profiles will inform, but likely exhibit some differences from, GA4GH Standard Profiles.
• Broader data exchange with external systems may require data transforms (which will be simplified by a common basis in the Core IM).

ga4gh.org

Goal is to bring the GK-Pilot work that includes Implementation Profiles for several VA Statement types from an initial alpha version to a beta version that is more closely aligned with the VA Core IM

VA Core IM Resources

• VA Core IM v0 source of truth: schemasheets format google sheet
• web docs: LinkML generated Github Pages
• Core IM v0 Lucidchart Diagram
• Starter Variant Pathogenicity Profile Lucidchart Diagram

GK Pilot Implementation Profile Resources

1. Profiling process Lucidchart diagrams
2. yaml schema documents
1. core source.yaml
2. vrs-source.yaml
3. annotation-source.yaml
4. varstatement-source.yaml
3. Variant Pathogenicity Statement ClinVar json data examples

​

​

Model Comparison Artifacts

• Variant Pathogenicity gk-pilot vs VA Starter Lucidchart Diagrams
• VariantPathogenicity Statement ClinVar examples, annotated with diffs from VA model (json file)
• Condition representation with Mappings vs Codeable Concepts json examples

​

​

GK-pilot <---> VA Core Alignment Exercise

ga4gh.org

• Implementation of the Proposition Assessment Paradigm
• Explicit SPOQ structure of the Statement's Proposition
1. attribute names
2. predicate requirement
3. qualifier structure/naming
• Modeling of direction and strength of an assessment (confidence/evidence)
• The Mapping class: utility, and comparison to CodeableConcept
• Inclusion of risk allele classifications in the VarPath Statement Profile
• Use cases for RecordMetadata class
• Coding and IRI bindings to certain attributes.
• enums vs Codings for value sets

​

GK-pilot <---> VA Core Alignment Questions

ga4gh.org