VRS AnVIL: Connecting VCF to Clinical Evidence

Brian Walsh

Quinn Wai Wong

Introduction

https://github.com/ga4gh/va-spec

https://github.com/ga4gh/vrs

https://github.com/cancervariants/metakb

Introduction

Subject VCF/VUS

Genotype and Phenotype Search tools

VRS AnVIL Toolkit

is a

Precomputed indices

Genomic Knowledge

Evidence Mapper

meta_kb

is a

civicdb.org

moalmanac.org

clinicalgenome.org*

normalizes

hits

Cohort Building

Downstream

Workflows

Presentation Outline

VRS AnVIL Toolkit as a solution

Existing tools to annotate VCFs with evidence

Proof of concept with 1000 Genomes

Usage, discussion, & future work

Existing Tools: Annotating VCFs with Evidence

The GA4GH Variant Representation specification has helped standardize the exchange of variant data (vrs-python)

• Schema to represent alleles, sequence references, sequence expressions, etc
• Consistent, unique global identifiers
• Minimal dependencies (seqrepo)

Docs

VRS enables fixed-length ID creation from a given genomic expression by means of a VRS object (vrs-python)

[Location]

​

​

[Prefix]

ga4gh:VA.

+

​

[Digest]

rRPCnh0XXjuePRGWerw6PhVXFYjhchwP

+

​

[State]

Computed Identifiers

The VICC MetaKB is a harmonized data warehouse for clinical variant interpretations (MetaKB)

• Consolidate variant data over minimally overlapping knowledgebases
• Publicly accessible API
• Uses and supports VRS IDs!

Wagner et al 2020

Solution: VRS AnVIL Toolkit

vrs_anvil_toolkit provides a CLI to retrieve clinical data from VCF file in a configurable fashion

• Define configurations
• Run CLI

vrs_anvil_toolkit combines variant translation and clinical interpretations into a single data collection workflow

1. Organize runtime settings in config file
1. Configs: performance, strictness
2. Directories and file paths
2. Translate variants to VRS IDs
3. Identify matches in the MetaKB cache
4. Write to logs and metrics file

vrs_anvil_toolkit combines variant translation and clinical interpretations into a single data collection workflow

• Organize runtime settings in config file
• Translate variants to VRS IDs
• Parse set of VCF files
• Multiple processes through CLI
• Identify matches in the MetaKB cache
• Write to logs and metrics file

nohup vrs_bulk annotate —scatter &

vrs_bulk ps

vrs_bulk annotate

vrs_anvil_toolkit combines variant translation and clinical interpretations into a single data collection workflow

• Organize runtime settings in config file
• Translate variants to allele VRS ID
• Identify matches in the MetaKB cache
• Writing to logs and metrics file

vrs_anvil_toolkit combines variant translation and clinical interpretations into a single data collection workflow

• Organize runtime settings in config file
• Translate variants to allele VRS ID
• Identify matches in the MetaKB cache
• Write to logs and metrics file
• Log: invalid alleles, errors, progress
• Metrics: successes, timing, matches by file

Proof of Concept : 1000 Genomes

About the 1000 Genomes Dataset

​

• AnVIL_1000G_PRIMED-data-model
• High coverage sequencing of samples
• 3202 samples (patients)
• 23 VCFs organized by chromosome

Using public 1000 Genomes Project data on Terra, we can consolidate cohort-level stats and evidence.

• Organizing runtime settings in config file
• Identifying variants matches to metakb cache using VRS IDs
• Organizing metrics into analysis-ready data
• Get sample-level variant data
• Aggregate metakb evidence
• Creating figures and evidence from data

Using 3202 samples from the 1000 Genomes Project on Terra, we can consolidate cohort-level stats and evidence

• Organizing runtime settings in config file
• Identifying variants matches to metakb cache using VRS IDs
• Organizing metrics into analysis-ready data
• Gather analyses from data
• Percent of patients w/ variant match
• Number of variants per patient
• Examples of evidence

Patients had successful variant ID matches to the CIViC knowledge but not to Molecular Almanac.

The spread varies between the germline and somatic-labelled MetaKB variants.

Even for doing cohort-level aggregation, MetaKB evidence is still accessible in the processed results.

Most common variant: 2952/3202 = 92.2%

(19-43551574-T-C) (ga4gh:VA.SPP7r7F_Wb3XbNY8Fawk91yt1U03eIVV) (civic.eid:673)

The XRCC1 R399Q variant was correlated with increased response to platinum-based neoadjuvant chemotherapy in patients with cervical cancer. Tumor samples from 36 patients with Stage IB or IIA bulky (greater than 4 cm in size) cervical carcinomas were used in this study.

Therapeutic Evidence for Top 2 VRS IDs

Usage, Discussion, & Future Work

Usage

vrs-anvil (private)

https://github.com/ohsu-comp-bio/vrs_anvil_toolkit

pip install vrs_anvil_toolkit

Depending on your use case, it might be helpful to use vrs-python and MetaKB individually.

• vrs-python: direct translation from the source, annotating VCF files with VRS IDs
• MetaKB: existing VRS ID data, one-off queries
• vrs_anvil_toolkit: performance (threading, caching, multiprocessing), Terra support, unified configuration, error handling

There’s seems to be a performance difference between Terra and other platforms

Throughput (vrs_bulk annotate)

• Terra: 850 variants/s
• Mac (M3): 4,600 variants/s

Pytest (test_gnomad)

• Terra: 10.6s
• Raw GCP: 3.0s
• Mac (M3): 2.9s

We want to continue to build out the toolkit’s functionality to support GREGoR and other dataset

Acknowledgements

Wagner Lab: Kori Kuzma

Ellrott Lab: Brian Walsh, Kyle Ellrott