ga4gh.org

GKS: The Schemarillion

​

​

Bob Freimuth

Aly Khalifa, Salem Bajjali, Sarah Senum, Jeff Chen

​

ga4gh.org

The Silmarillion

​

J. R. R. Tolkien

ga4gh.org

Table of Contents

Preface: Of the Beginning 5

Chapter 1: Of HL7 and FHIR 10

Chapter 2: Of Modeling 20

Chapter 3: Of the Coming of GKS into the FHIR 40

Chapter 4: Of the Ruin of Interoperability 60

Chapter 5: Of the Schemas of Power and the Next Age 80

​

ga4gh.org

Strategic Goals

• Maximize both semantic expressiveness and computability of genomic data
• Maximize interoperability of genomic data across the research-translational-clinical spectrum
• Enable seamless integration of public knowledge bases on GA4GH with clinical systems on FHIR

ga4gh.org

Tactical Goals

Main Objective

• Interoperable genetic data for clinical and research purposes

Initiatives

• Develop and encourage alignment between HL7 CG and GA4GH to represent genomic data
• Develop formal mappings between HL7 FHIR and GA4GH GKS that could be implemented as transformative tooling

6

ga4gh.org

Of HL7 and FHIR

HL7 Standards

• Transmission of health-related information via messages
• Patient records
• Lab results
• Billing information

​

• U.S. regulatory policy mandates support for HL7 standards
• EHR vendors
• Lab information systems

​

https://www.ringholm.com/docs/the_early_history_of_health_level_7_HL7.htm

©2018 MFMER | slide-8

HL7 Standards

• Version 2 messaging (V2) – since 1989, still in wide use
• Version 3 messaging (V3) – not adopted
• Clinical Document Architecture (CDA)
• Continuity of Care Document (CCD)
• Structured Product Labeling (SPL)
• Arden Syntax – recommendations as Medical Logic Module
• Fast Healthcare Interoperability Resources (FHIR) – new(est)!

http://www.hl7.org/implement/standards/ansiapproved.cfm

©2018 MFMER | slide-9

HL7 V2 Messaging

• V2.x
• 2.1, 2.2, … 2.8.2
• Message structure
• Line-based (“segments”)
• Delimited
• Fields: pipe “|”
• Components: carat “^”
• Subcomponents: ampersand “&”
• Others…

©2018 MFMER | slide-10

HL7 FHIR®:�Fast Healthcare Interoperability Resources

• Goal: facilitate implementation
• Not semantic expressivity!
• Standard internet technologies
• HTTP, REST, SSL, Oauth
• XML, JSON, RDF
• Improved portability
• e.g., EHRs, mobile devices
• Supported by many vendors
• Interoperability through predefined resources and APIs

​

​

www.hl7.org/fhir

©2018 MFMER | slide-11

Fast Healthcare Interoperability Resources (FHIR)

• History
• Sept 2011 – initiated
• Feb 2014 – R1 as DSTU
• Oct 2015 – R2
• Mar 2017 – R3
• Oct 2019 – R4 as normative
• Mar 2023 – R5 as STU
• ? – R6 (as normative??)

​

©2018 MFMER | slide-12

HL7 FHIR®: Fast Healthcare Interoperability Resources

• Resources
• Represent clinical concepts
• Content model (including structure)

DiagnosticReport

Resource

Observation

ServiceRequest

©2018 MFMER | slide-13

HL7 FHIR®: Fast Healthcare Interoperability Resources

• Resources
• Represent clinical concepts
• Content model (including structure)
• Extended and specialized via Profiles

DiagnosticReport

Genetics Diagnostic Report

Resource

Profile

Observation

Genetic Observation

ServiceRequest

Genetic Test Order

©2018 MFMER | slide-14

HL7 FHIR®: Fast Healthcare Interoperability Resources

• Resources
• Represent clinical concepts
• Content model (including structure)
• Extended and specialized via Profiles
• Resources and Profiles are building blocks
• Target the most common "80%"
• Domain experts inform the design

DiagnosticReport

Genetics Diagnostic Report

Resource

Profile

Observation

Genetic Observation

ServiceRequest

Genetic Test Order

©2018 MFMER | slide-15

HL7 FHIR®: Fast Healthcare Interoperability Resources

• Resources
• Represent clinical concepts
• Content model (including structure)
• Extended and specialized via Profiles
• Resources and Profiles are building blocks
• Target the most common "80%"
• Domain experts inform the design
• Implementation Guides are compositional
• Collection of Profiles

DiagnosticReport

Genetics Diagnostic Report

Resource

Profile

Observation

Genetic Observation

ServiceRequest

Genetic Test Order

Implementation Guide

©2018 MFMER | slide-16

FHIR Resources

https://build.fhir.org/modules.html

Levels 1 & 2: Infrastructure

Levels 3 & 4: Content

Level 5: Reasoning

©2018 MFMER | slide-17

Profile of Observation: Patient Blood Pressure

• Code is set to LOINC 85354-9 (“blood pressure panel”)
• Value is constrained out (cardinality is 0..0)

©2018 MFMER | slide-18

https://build.fhir.org/datatypes.html

ga4gh.org

ga4gh.org

ga4gh.org

Genomics Reporting Implementation Guide

https://registry.fhir.org/guides

http://hl7.org/fhir/uv/genomics-reporting/index.html

​

​

​

​

​

http://build.fhir.org/ig/HL7/genomics-reporting/

​

Genomics Reporting IG: Scope

• Representation of genomic variants, including simple discrete variants, structural variants including copy number variants, complex variants as well as gross variations such as extra or missing chromosomes
• Both known variants and fully describing de novo variations
• Germline and somatic variations
• Mitochondrial DNA variations
• Mosaicism
• Relevance of variations from the perspective of disease pathology, pharmacogenomics, transplant suitability, etc.
• Any technology/assay method: full and partial DNA sequencing, including whole genome and exome studies

​

​

​

CG Reporting Implementation Guide (STU2, R4)

CG Reporting Implementation Guide

Genomic Observations: Profiles of Profiles of Profiles

Genomic Findings: Relationships Between Profiles

AMIA 2021 Annual Symposium | amia.org

"Big bag of fields"

• Genomic Location components
• Variant Location components
• Variant Change components
• Reference sequence
• Amino Acid Change components
• Allelic State components

Profile(s) of Observation + Components: Lessons Learned

Pros

• Discrete genomic results!!
• (Mostly)
• Uses existing normative FHIR resource (Observation)
• Low barrier to implementation
• Similar approach to existing v2 messages (key-value pairs)

​

Cons

• Data structures do not reflect the semantics of the domain
• Flat list
• Related attributes are not associated to each other
• Requires detailed understanding of definitions and intent to implement consistently
• Still contains semantic ambiguity (especially for complex structures)

​

Of Modeling

HL7 Clinical Genomics Information Modeling

A subgroup of the CG WG will be formed to focus on the development of an information model to represent the clinical genomics domain. The creation of this model is consistent with current CG WG project scope statements and it is a necessary component of the future consolidated standard, which will include the current Domain Analysis Model and Domain Information Model, as described in the CG WG DAM PSS (Q1/2016).

​

• A model is needed to unify our specifications

​

CG Information Model (IM)

• Information Model
• Technology agnostic
• Focused on the data
• Not clinical processes
• Not assay/testing technology
• Not bioinformatic analytical pipelines
• Separate from interpretations
• Captures semantics
• Concepts, relationships
• Promote harmonization and inter-artifact consistency among all CG WG standards

​

• Builds on existing work
• Reuse when possible
• Extend/remodel when necessary
• Models, ontologies, implementations (FHIR, v2)
• But an IM is not an implementation
• Aligns to other work
• GA4GH
• ClinGen
• CIVIC
• etc

COMPLEX USE CASES FOR GENOMICS TESTING

• Trio testing, family studies
• Multiple WES/WGS (>2 subjects)
• Secondary findings/analysis
• Tumor:Normal testing
• Serial tumor sequencing
• Same “normal” reference
• Cascade/reflex testing
• Observed results produced by different assay technologies
• “Negative” tests are still informative
• Need to know regions assayed and technologies used even if no variants are reported

• Variant reinterpretation
• New knowledge, regardless of report or assay
• New test result, in context of multiple other variants from previous tests
• Need access to non-reported results
• Result reconciliation
• Screening => diagnostic
• Targeted <=> WES/WGS
• Risk score calculation
• Pharmacogenomics: drug selection or dosing algorithms
• Polygenic risk scores

©2023 Mayo Foundation for Medical Education and Research | slide-33

HL7 FHIR GENOMICS (R5+)

Genomic Study

Genomic Analysis

Part of,

Derived from

Genomic Diagnostic Report

(profile)

Observed Variation (profile)

Quality

Confidence

ACMG classification

AMP tier

See Mol Var models

Called variation*

Clinically reported variation

Molecular Variation

See Mol Var models

Molecular Sequence

See Mol Seq model

Not shown: implication, interpretation, etc

Called sequence*

Definitional,

Patient/Obs-agnostic

(e.g., knowledgebase)

Observed Sequence (profile)

Quality

Confidence

See Mol Seq model

Under Development

Observational,

Patient-Specific

GA4GH-Aligned

GA4GH-Aligned

* Called variation and sequence are not restricted to clinical testing and reporting (e.g., research studies)

©2023 Mayo Foundation for Medical Education and Research | slide-34

HL7-GA4GH Alignment: Vision

GA4GH VRS: definitional, computational variation

GA4GH VA: structured variation annotation

Knowledge bases

Genetic reports

EHRs

HL7 FHIR Genomics IG: clinically observed & reported variation

Aligned information models (and technical schemas)

©2018 MFMER | slide-35

Molecular Sequence

CG Information Model:

Core Sequence Classes

Definition of Sequence:

• Contiguous primary sequence (no gaps)
• Any type (DNA, RNA, protein)
• Not contextualized within another Sequence

Transformations between SequenceRepresentations

• Do not change content/meaning (Sequence)
• Do change representation (serialization)

Support complex recursion

Literal

Formatted

Resolvable

Extracted

Repeated

Concatenated

Relative

ACGTACGT

Transformations between SequenceRepresentations

• Do not change content/meaning (Sequence)
• Do change representation (serialization)

Literal

Formatted

Resolvable

Extracted

Repeated

Concatenated

Relative

ACGTACGT

>FASTA header

ACGTACGT

<URI> to retrieve an instance of Sequence

AAAAACGTACGTTTTT

ACGT[2]

CCGTACGT + Edit[ C>A at position 0 ]

A + Formatted[CG] + Extracted[TA] + Resolvable[CGT]

Transformations between SequenceRepresentations

• Do not change content/meaning (Sequence)
• Do change representation (serialization)

Literal

Formatted

Resolvable

Extracted

Repeated

Concatenated

Relative

ACGTACGT

>FASTA header

ACGTACGT

<URI> to retrieve an instance of Sequence

AAAAACGTACGTTTTT

ACGT[2]

CCGTACGT + Edit[ C>A at position 0 ]

A + Formatted[CG] + Extracted[TA] + Resolvable[CGT]

Supported by VRS

Molecular Sequence

Identifier, type

Sequence Representations

https://build.fhir.org/branches/cg-im-molseq-work_in_progress_2/molecularsequence.html

Supporting Datatypes: Molecular Location

Location = Context + Interval

Context

Sequence

GenomeAssembly

Interval

Coordinate Cytoband

©2023 Mayo Foundation for Medical Education and Research | slide-42

Molecular State

State might also need to be defined by a class of sequences (residue ambiguity or representative/ reference sequence)

Loci can be coordinate-based or feature-based

Allele

-State 1..1 Ref(MolecularSequence)

-Locus 1..1 SequenceLocation

Haplotype

-State 1..* Allele in cis

-Locus 1..1 SequenceLocation

Genotype

-State 1..* Allele | Haplotype in cis or trans

-Locus 1..1 SequenceLocation (homologous)

Generalized Conceptual Model

-State 1..* Ref(MS) | Allele | Haplotype

-Locus 1..1 SequenceLocation

​

​

​

Generalization, Abstraction, and FHIR-ization

MolecularState Resource

-memberState 0..* Ref(MolecularState)

-locus 0..* MolecularLocation

Profiling

Sequence

-State 1..1 Ref(MolecularSequence)

-Locus 0..0 SequenceLocation

Simplified Profile Structure

Molecular State

(Resource)

Allele

(Profile)

Sequence

(Profile)

Haplotype

(Profile)

Genotype

(Profile)

• Specialized FHIR resource(s) for structured omic data
• Definitional (does not require an Observation)
• Can create new profiles for new types of omic data

Genomics Reporting IG (R5)

CG Information Model:

Core Sequence Classes

Domain concept (data type)

Representations for instance data (dependent on use case)

©2023 Mayo Foundation for Medical Education and Research | slide-48

MolecularState

type [0..1]

identifier [0..*]

name [0..*]

locus [0..*] MolLocusDT

Representation

Literal

encoding

value

Formatted

file

Resolvable

uri

format

Extracted

startingSeq [1..1] MolState

interval�isRevCompd

Repeated

seqMotif [1..1] MolState

copyCount

Concatenated

element [1..*] (ordered) MolState

Relative

startingElement [1..1] MolState

Edit

interval

replacedSeq [0..1] MolState

replacementSeq [1..1] MolState

0..*

1..* ordered

MemberGroup

members [1..*] MolState

MolecularState

Base resource

©2023 Mayo Foundation for Medical Education and Research | slide-49

Molecular Sequence: Simple Representations

MolSeq

.type code: “genomic DNA”

.identifier ID: “internal seq 123”

.name str: “MFG genomic sequence”

.literal

.sequenceValue str: “ACGTACGT”

.encoding CC: “IUPAC, ACGT only”

.formatted (file as Attachment)

.contentType code: MIME type for “FASTA”

.data base64binary: “...”

.formatted (resolvable as Attachment)

.contentType code: MIME type for “FASTA”

.url “http...”

.repeated

.sequenceMotif Ref(MolSeq)

.copyCount int: “2”

Core of a MolSeq instance

Plus 0..* Representation instance(s):

AAAAACGTACGTTTTT

ACGT

Molecular Sequence: Concatenated Representation

MolSeq

.type code: “genomic DNA”

.identifier ID: “internal seq 123”

.name str: “MFG genomic sequence”

Core of a MolSeq instance

.concatenated

.element

.element Ref(MolSeq)

​

​

​

​

​

​

​

.ordinalIndex int: 1

MolSeq

.type code: “genomic DNA”

.literal

.sequenceValue str: “A”

.encoding CC: “IUPAC, ACGT only”

.element

.element Ref(MolSeq)

​

​

​

​

​

​

​

.ordinalIndex int: 2

MolSeq

.type code: “genomic DNA”

.formatted (file as Attachment)

.contentType code: “FASTA”

.data base64binary: “...”

.element

.element Ref(MolSeq)

​

​

​

​

​

​

​

.ordinalIndex int: 3

MolSeq

.type code: “genomic DNA”

.formatted (resolvable as Attachment)

.contentType code: “FASTA”

.url “http...”

Plus 0..* Representation instance(s):

This representation contains 3 elements, each of which defines a sequence and uses different representations themselves. The 3 elements are concatenated together in order to result in “ACGTACGT”

“CG”

“TACGT”

“A”

Molecular Sequence: Relative Representation

MolSeq

.type code: “genomic DNA”

.identifier ID: “internal seq 789”

.name str: “MFG2 genomic sequence”

Core of a MolSeq instance

Plus 0..* Representation instance(s):

.relative

.coordinateSystem code for “0-based interbase”

.startingSequence

edit

.editOrder 0

.start 1

.end 2

.replacedSeq Ref(MolSeq.literal=“C”)

.replacementSeq Ref(MolSeq.literal=“T”)

edit

.editOrder 1

.start 4

.end 5

.replacedSeq Ref(MolSeq.literal=“A”)

.replacementSeq Ref(MolSeq.literal=“G”)

MolSeq

.type code: “genomic DNA”

.identifier ID: “internal seq 123”

.name str: “MFG1 genomic sequence”

Edits ATGTACGT to ATGTGCGT

Edits ACGTACGT to ATGTACGT

MFG1: ACGTACGT

(seq rep not shown)

MFG2: ATGTGCGT

Molecular Haplotype: Examples of Representations

MolHaplotype

.type code: “genomic DNA”

.identifier ID: “MFG_0001.A”

.name str: “MFG*1A”

.locus SeqLocDT

.formatted (file as Attachment)

.contentType code: MIME type for ...

.data base64binary: “...”

.formatted (resolvable as Attachment)

.contentType code: MIME type for ...

.url “http...”

Core of a MolHaplotype instance

Plus 0..* Representation instance(s):

.locus is currently optional because it could be determined from certain Representations (should it be mandatory?)

.memberGroup

.members [Ref(Allele_1), Ref(Allele_2)]

.memberGroup is probably the most intuitive and might be the most often used Representation for Haplotype

.formatted is provided to support file-based conventions (not common now, but likely in the future)

Molecular Genotype: Examples of Representations

MolGenotype

.type code: “genomic DNA”

.identifier

.name

.locus SeqLocDT

.formatted (file as Attachment)

.contentType code: MIME type for “VCF”

.data base64binary: “...”

.formatted (resolvable as Attachment)

.contentType code: MIME type for “VCF”

.url “http...”

Core of a MolGenotype instance

Plus 0..* Representation instance(s):

.locus is currently optional because it could be determined from certain Representations (should it be mandatory?)

.memberGroup

.members [Ref(Allele_1), Ref(Allele_2)]

.memberGroup is probably the most intuitive and might be the most often used Representation for Genotype (note either Allele or Haplotype can be used as members)

.literal

.sequenceValue str: “G/T”

.encoding CC: “IUPAC, ACGT only”

.memberGroup

.members [Ref(Haplo_1), Ref(Haplo_2)]

Transformations Between Entities

Sequence

Genotype

Haplotype

Allele

+ Seq context (LocusDT)

- Seq context

group

ungroup

ungroup

group

ungroup

group

Not shown: transformations among Representations for a given entity

(including normalizing to a simple/literal value that can be used to test for equivalency)

More Omic Data Types

https://fusions.cancervariants.org/en/latest/introduction.html

Fusion Nomenclature

Gene

HGNC symbol 1..1 Str

HGNC id 1..1 ID

Transcript Segment

Transcr seq id 1..1 ID

5’ segment boundary 0..1 SegB

3’ segment boundary 0..1 SegB

Linker Seq

Seq 1..1 Str

https://fusions.cancervariants.org/en/latest/introduction.html

Structural Elements

Templated Linker Seq

Genomic Loc 1..1 GenLoc

Genomic strand 1..1 CC

Seq 0..1 Str

CC

MolSequence.id

MolSequence.literal

Missing concept of “linker” (role of the seq)

SeqLoc datatype

Sequence Location

Ref seq id 1..1 ID

startCoord 1..1 Int

endCoord 1..1 Int

Genomic Location

A SeqLoc where ref seq is chrom ref in a genome assembly

Segment Boundary (SegB)

Exon number 1..1 Int

Exon offset 1..1 Int

Genomic location 1..1 GenLoc

Seq Feature

SeqLoc datatype

Missing concept of “offset” for coord interval

Missing concept

Fusion Nomenclature

https://fusions.cancervariants.org/en/latest/introduction.html

Regulatory Feature

Regulatory class 1..1 CC

Feature ID 0..1 ID

Feature Loc 0..1 GenLoc

Associated Gene 0..1 Gene

Regulatory Elements

Functional Domain

Label 0..1 Str

id 0..1 ID

Seq Loc 0..1 SeqLoc

Status 1..1 CC

Associated Gene 1..1 Gene

Reading Frame

isPreserved 0..1 Bool

Categorical Elements

Seq Feature

SeqLoc datatype

CC

Seq Feature

SeqLoc datatype

CC

CC

Missing concept. Is this at the level of an element or overall transcript? Is this transcript only or also gene?

Causative Event

Type 1..1 CC

Descr 0..1 Str

Assay

Name 1..1 Str

Id 1..1 CC

Fusion detection 1..1 CC

Method URI 1..1 URI

Assayed Elements

ActivityDef, PlanDef, Task, Procedure, CC, …

Missing concept

Fusions as a type of MolecularState

MolecularState

type [0..1]

identifier [0..*]

name [0..*]

locus [0..*] MolLocusDT

Representation

0..*

Text-based (unstructured or formal grammar)

Ordered series of structured fusion components

©2018 MFMER | slide-60

Molecular State: Modeling TBD

Copy Number

• Is this simply Locus + Count?

Structural “variation”

• Describes a state, not a change (?)
• Several representations
• ISCN, CIVIC, HGVS
• VCF, GA4GH
• Genome graph

​

​

​

ga4gh.org

Annotations

Sequence and Allele:

• Feature

Allele, Haplotype, Genotype, Copy Number, Structural “Variation”:

• Association (e.g., disease)
• Function
• Other?

​

How much structure is needed in knowledge statements?

​

ga4gh.org

SequenceFeature

​

identifier: Identifier [0..*]

type: CodeableConcept [1..1]

name: string [0..*]

contexts: SequenceContext [0..*]^B

Sequence

​

type: CodeableConcept [1..1]

identifier: Identifier [0..*]

name: string [0..*]

representation: SequenceRepresentation [0..*]

features: LocatedFeature [0..*]^A

LocatedFeature

​

feature: SequenceFeature [1..1]

location: Location [1..1]

strand: CodeableConcept [0..1]

Draft Transcript Model

Logical Model

Transcript

​

sequence: Sequence [1..1]

exons: LocatedFeature [0..*]

codingRegion: LocatedFeature [0..*]

features: LocatedFeature [0..*]

associatedGene: Identifier [0..*]

Notes:

• Sequence.type may be constrained to codes for cDNA, RNA, mRNA, etc
• SequenceFeature.type is used to codify the feature using terms (e.g., from an ontology). For example, SF.type could be a code for "exon" or "coding region".
• Transcript.associatedGene is added for convenience. It may also be desirable to add this attribute to SequenceFeature (it was not included in the conceptual models) to support features from different genes (e.g., fusion events).

Draft Sequence Annotation Model (for reference)

VRS Schema

SequenceContext

​

sequence: Sequence [1..1]

location: Location [1..1]

strand: CodeableConcept [0..1]

^A Used only for Sequence-based structures

^B Used only for Feature-based structures

ga4gh.org

A Generic VA Statement Model

Statement Semantics are captured by four dedicated elements:

• subject: the entity describe by the statement (a variation)
• predicate: the relationship asserted to hold between the subject and object
• object: an entity/concept associated with the subject
• qualifier(s): refine or extend the meaning of the core S-P-O triple

Example: EGFR-L858R (subject) predicts_sensitivity_to (predicate) Afatinib (object) in NSCLC (qualifier)

​

​

​

​

​

This pattern makes it clear which elements of a statement comprise its core meaning - so it is clear to human and computational agents what is being asserted as true

Of the Coming of GKS into the FHIR

The Best of Both Worlds

VRS

• Computable representations
• Minimalistic, no optional fields
• Agnostic of use case

FHIR

• Must support unstructured text
• Very complex but extensible
• Defines clinical context and use

​

©2018 MFMER | slide-66

Conceptual Mappings: Variation (State)

67

FHIR

VRS

Conceptual Mappings: Sequence Expression

68

FHIR

VRS

Logical Mappings: �Sequence Expressions

69

FHIR

VRS

X

X

X

Logical Mappings: Literal Sequence

70

FHIR

VRS

X

String that matches

^[A-Z*\-]*$

and is derived from IUPAC

Logical Mappings: Derived/Extracted Sequence

71

FHIR

VRS

-sequence_id

-interval

VRS requires 0-based interbase coordinate system

X

X

Logical Mappings: Derived/Extracted Sequence

72

FHIR

VRS

Additional MolecularSequence Instance

VRS requires 0-based interbase coordinate system

Logical Mappings: Derived/Extracted Sequence

73

FHIR

VRS

VRS requires 0-based interbase coordinate system

Logical Mappings: Repeated Sequence

74

FHIR

VRS

X

X

Logical Mappings: Repeated Sequence

75

FHIR

VRS

X

Additional MolecularSequence Instance

Nested SequenceExpression

Logical Mappings: Repeated Sequence

76

FHIR

VRS

X

Additional MolecularSequence Instance

Nested SequenceExpression

FHIR Range and Quantity Datatypes

77

https://www.hl7.org/fhir/datatypes.html#Range

https://www.hl7.org/fhir/datatypes.html#Quantity

​

Logical Mappings: Composed/Concatenated Sequence

78

FHIR

VRS

VRS assumes an ordered list 

Logical Mappings: Composed/Concatenated Sequence

79

FHIR

VRS

VRS assumes an ordered list 

3X

Repeated for each components

Logical Mappings: Composed/ Concatenated Sequence

80

FHIR

VRS

VRS assumes an ordered list 

3X

Conceptual Mappings: Location Datatype

VRS

CytobandLocation

.genomeAssembly

​

​

​

​

​

​

.cytobandInterval

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

GenomeAssembly

.organism “Homo sapiens”

.build “GRCh38.p14”

.accession “NC_000002.12”

.description

CytobandInterval

.chromosome “2”

.start

​

​

​

​

​

​

.end

​

​

​

​

​

​

Cytoband

.arm “q”

.region “2”

.band “1”

.subband “1”

Cytoband

.arm “q”

.region “2”

.band “1”

.subband “2”

MolecularLocation

Conceptual Mappings: Location Datatype

82

FHIR

VRS

MolecularLocation [Datatype]

X

X

X

X

Of the Ruin of Interoperability

GA4GH-HL7 Alignment: Gaps in FHIR R5+

• Models still under development
• Probably OK:
• Copy number, Haplotype, Genotype
• Possible gap in scope:
• Variation set (not in scope for FHIR?)
• Possible difference in semantic/structural model:
• Gene (FHIR sequence feature)

​

​

• Location Datatype does not yet support imprecise intervals (VRS IndefiniteRange)
• Support via FHIR datatypes Range and Quantity?

​

GA4GH-HL7 Alignment: Gaps in VRS

• Support for:
• Sequence name, ID, type (also impacts Haplotype, Genotype)
• Cytoband interval (discrete)
• Recursion in Sequence Representations (?)

​

​

​

​

​

• Possible perma-gaps:
• Encoding for literal sequence
• Coordinate numbering systems
• Formatted, Resolvable, and Relative sequence representations

​

Of the Schemas of Power and the Next Age

Take-Home Points

• Common goals: computability, semantic rigor
• Different use cases, implementation philosophies and technologies
• Good alignment between GKS and FHIR R5+
• But some gaps exist
• Parallel development in progress
• Opportunity to converge
• Risk of divergence and siloing

​

ga4gh.org

Possible Immediate Next Steps

• Discuss gaps, intentionally decide if/how to bridge them

​

• Encourage continued convergence and harmonization
• Specification development and de novo modeling
• Critical review and constructive feedback

ga4gh.org

Opportunities for Harmonization

• Haplotype and Genotype Representation(s)
• Ensure at least one lossless touch point across specifications
• Models under development in both GA4GH and HL7 CG
• Fusions
• Sequence annotations/features
• Variation annotation (structured knowledge, state vs. change)
• Extension mechanism(s)
• User-defined fields
• New representations of core concepts (e.g., Allele, Genotype)
• New types of molecular data

89

ga4gh.org

Possible Future Activities

• Formal mappings between GA4GH and HL7 FHIR specifications

​

• Develop tooling to demonstrate GA4GH-FHIR transformations

​

ga4gh.org