1 of 35

ga4gh.org

2 of 35

GKS: The Two Toolkits

Andreas Prlić & Larry Babb (and Alex Wagner)

ga4gh.org

3 of 35

Agenda (draft)

Introduction

Biocommons Toolkit

VRS-Python Toolkit

Performance requirements, integrations with seq collections, etc..

Seq collections - how does seqrepo support or not ? (Andy)

ga4gh.org

4 of 35

Biocommons Toolkit

ga4gh.org

5 of 35

BioCommons Hackathon 2023 - SeqRepo

Manually updated SeqRepo to 2023-09-18
Script for autoupdating SeqRepo
Open PR: https://github.com/biocommons/biocommons.seqrepo/pull/124
Documentation updates!

ga4gh.org

6 of 35

BioCommons Hackathon 2023 - UTA

Dockerizing

https://github.com/biocommons/uta/pull/248/files#diff-96863ed3fda57ca744a4170035a58d91e0bad8f6c69c0b41f9f2d8d7fb521ce8
docker-compose.yml with postgres and python containers
postgres container loads latest version of UTA

Documentation updates!

It’s possible to do developer setup on MacOS now

ga4gh.org

7 of 35

BioCommons Hackathon 2023 - UTA

Alignment

Avoid using uta-align to re-align every exon
Now use 2 new files from NCBI (that have not been available previously)
Work on re-building the exon-alignment and CDS start/stop merge

We are here!

This is the goal!

ga4gh.org

8 of 35

BioCommons Hackathon 2023 - UTA

GitHub Action

ga4gh.org

9 of 35

BioCommons Hackathon 2023 - AnyVar

Added vrs-python translator backend to anyvar
Improved anyvar setup documentation
Added definite CNV HGVS translation to vrs-python
Added definite CNV HGVS storage in anyvar
Updated anyvar object models to vrs-python 2.0
Identified a bug in ga4gh_digest function
Added support for definite/indefinite range intervals in hgvs
Added support for unbalanced range intervals in hgvs

ga4gh.org

10 of 35

BioCommons Hackathon 2023 - HGVS Parsing

ga4gh.org

11 of 35

BioCommons Hackathon 2023

ga4gh.org

12 of 35

High Throughput Sequence Retrieval and Alignment Operations

ga4gh.org

13 of 35

Public dataset sizes

If it takes ~ 1 hour to normalize/register all variants in a full WGS VCF, that’s too slow for a high throughput lab.

ga4gh.org

14 of 35

Refget and sequence collections - a refresher

Refget v2

GA4GH Standard to access reference sequences & metadata

Generates unique identifiers based on sequence content

Supports common unique identifiers such as those from public sequence repositories

Sequence Collections

Collate groups of sequences

Same identifier generation code to build a single identifier

Uses an array storage system to simplify checksum generation and separate properties

Applies to genomes, proteomes, transcriptomes

ga4gh.org

15 of 35

Refget v2 and seqrepo REST

What changed in v2 of refget (the latest approved version)

Support namespaced identifiers preferring CURIE-style IDs (similar to seqrepo)

refseq:NP_001274413.1
ga4gh:SQ.aKF498dAxcJAqme6QYQ7EZ07-fiw8Kw2

md5 and sequence sha512t24u identifiers can be passed through without namespaces
sha512t24u the preferred checksum algorithm

MD5 is still encouraged for CRAM compatibility

Adopting GA4GH /service-info

ga4gh.org

16 of 35

Sequence collections - the structure

JSON object where each sequence collection attribute is a property of the JSON

ADR 2021-06-30

ga4gh.org

17 of 35

Sequence collections

JSON object where each sequence collection attribute is a property of the JSON

ADR 2021-06-30

ga4gh.org

18 of 35

Computing sequence identifiers

General algorithm

Normalisation: convert JSON into unique string representation

Using RFC8785: JSON Canonicalization Scheme (JCS)

Hashing: SHA512 truncated to the first 24 bits and converted to base64

seqcol

Normalisation

Hashing

digest

ADR 2023-01-12

ga4gh.org

19 of 35

Sequence collections - pangenomes

Pangenome & T2T-CHM13 demo

When we have these, we can digest and make identifiers for pangenomes

Non-collated & non-inherent

ga4gh.org

20 of 35

Sequence collections - next steps

Areas to be finalised with the spec

The standard schema accompanying collections
Naming around key fields
Documentation - https://seqcol.readthedocs.io/en/dev/specification/

Other areas to be finalised

Adoption of sequence collections
Identified EVA (Tim Cezard) and Ensembl (Andy Yates) as potential implementers both at EMBL-EBI
Looking for other adopters beyond EMBL-EBI

ga4gh.org

21 of 35

Cool Seq Tool

Common Operations On Lots-of Sequences Tool
Data Sources

SeqRepo: Fast access to sequence data
UTA: Transcript alignment data
PyLiftover: Liftover between assemblies (GRCh37 ←→ GRCh38)

Super Cool Functions

Alignment mapper (p → c → g)
Liftover to preferred MANE compatible transcript
Genomic coordinates ← → Transcript exon

ga4gh.org

22 of 35

VRS-Python Toolkit

ga4gh.org

23 of 35

VRS sounds great but…

ga4gh.org

24 of 35

gnomAD VRS pilot: VCF-Annotator pipeline

VCF annotation with VRS Ref and Alt IDs in the info field of a VCF using VRS-Python.

Annotates ~1,211 alleles/second

Worked with gnomAD team to add VRS into v3 hail tables

ga4gh.org

25 of 35

VRS v1.3 at scale: gnomAD

Added VRS for all 759.3 million variants in gnomAD v3 Hail Tables, plus reference alleles.�Over 1.5 billion VRS v1.3 Alleles, now searchable by range and computed identifier.

ga4gh.org

26 of 35

VRS population frequency

ga4gh.org

27 of 35

gnomAD Hail Utils to retrieve VRS population Frequency Statements

gnomAD utilities maintained by TGG @ Broad Institute:

https://github.com/broadinstitute/gnomad_methods/�Example notebook:�https://github.com/theferrit32/gnomad_methods/blob/gks-example-notebook/notebooks/gnomad-gks-v1.ipynb

New function gnomad_gks will annotate a range of variants with VRS Variation and VA frequency structures

Returned structure:

{

"locus": <hail locus>,

"alleles": <[ref .. alts]>,

"gks_vrs_variant": <VRS Variation>,

"gks_va_freq": <VA CohortAlleleFrequency Statement>

}

ga4gh.org

28 of 35

MaveDB: generalizing the SeqRepo model to map sequences

GA4GH sequence digests for MaveDB target sequences are registered with SeqRepo

Downstream VRS-Python methods can be utilized to construct normalized VRS alleles and haplotypes

ga4gh.org

29 of 35

Plans and experience from active project teams

ga4gh.org

30 of 35

Epic’s experiences with community tools

Using VRS to match patient variants against knowledgebases

Using homegrown VRS classes and HGVS parser for compatibility with languages and databases used by our infrastructure

hg19 VCF

hg38 VCF

VRS-formatted

patient variants

VRS-formatted

knowledgebases

variant matching

ga4gh.org

Epic is using VRS to match patient variants against knowledgebases during variant interpretation and reinterpretation workflows. Currently, we receive variants from a number of sources (hg19 VCF, hg38 VCF, HL7 v2 messages) and try to standardize their representation in VRS format. We also liftover hg19 variants and perform normalization on them.
We then plan to match these standardized patient variants against VRS-formatted knowledgebases. Alex, Larry and their teams are working on representing knowledgebases like ClinVar, gnomAD and CIViC in VRS format.
We are using a homegrown implementation of the VRS classes, rather than VRS-python. Part of the reason for this is that the database format we’re working with doesn’t support union types. For example, in VRS 1.3, an Allele can have a state of either a LiteralSequenceExpression, DerivedSequenceExpression or RepeatSequenceExpression, but there’s no direct way of representing this in our database. Instead, we have a flattened representation that combines all three classes but only has certain properties populated depending on the type. The other reason we are using our own VRS classes is because we work a lot in C#, and there was no existing VRS implementation in C#
Similarly, we created our own hgvs parser based on the rules listed on the hgvs website. This was mainly for language compatibility, but we also observed there were a few hgvs features that weren’t supported by the existing biocommons hgvs library, like imprecise coordinates on deletions and repeated sequences.

31 of 35

Epic’s experiences with community tools

Using pyliftover and bioutils.normalize for liftover and normalization

Works well for ~95% of cases
A single location in hg19 can liftover to multiple regions in hg38
Edge cases with changes in reference alleles

C

123

125

C

123

126

C

123

126

C

hg19 chr1

hg38 chr1

OR

ga4gh.org

32 of 35

VarCat

ga4gh.org

33 of 35

Thank you

ga4gh.org

34 of 35

ga4gh.org

35 of 35

ga4gh.org