Skip to content

SeqArray

Big-data storage and analysis of whole-genome sequencing variant calls in R/Bioconductor — an array-oriented, compressed GDS-backed container for genotypes and annotations from population-scale WGS, designed to scale to biobank-class cohorts where VCF stops being usable.

Talk at GBCC2025

  • Speakers: Xiuwen Zheng, Raven Chen
  • Day / session: Day 2 (Tue Jun 24, 2025) — Oral Session 2B, Grace Auditorium (chair: Vincent Carey)
  • Talk title: "Enhancing rare variant analysis in biobanks: whole-genome annotation using GDS files"
  • Slides: TBD
  • Video: GBCC2025 playlist
  • Abstract / schedule: GBCC2025 program

What it does

SeqArray builds on the GDS (Genomic Data Structure) format — a hierarchical, array-oriented, compressed on-disk container — to store genotype matrices, allele frequencies, and per-variant annotations from WGS at biobank scale. Access is through an R interface with C++-level performance, and the data layout is designed so that filtering by sample, variant, or annotation is fast without re-streaming the whole file. The GBCC2025 talk extends this with workflows for whole-genome annotation during rare-variant analysis — folding functional annotation directly into the GDS container so that downstream rare-variant tests (in SeqVarTools, SNPRelate, GENESIS) can access annotations without a separate join.

How it works

Core idea. GDS is "a flexible and scalable data container with a hierarchical structure that can store multiple array-oriented data sets." Genotypes are compressed to fewer than 8 bits per diploid call, and the hierarchical layout supports random access — so per-sample / per-variant slicing happens without scanning the whole file. The C++ CoreArray library does the work; the gdsfmt package exposes it to R.

Inputs / outputs. Inputs are VCF or BCF files converted with seqVCF2GDS(). Outputs are SeqVarGDSClass objects against which filtered queries are run; extracted data (genotypes, allele frequencies, INFO/FORMAT fields) come back as standard R objects.

Key innovation. Random-access, compressed, on-disk storage that scales past available RAM — the same storage substrate underlies SeqVarTools, SNPRelate, and GENESIS, which is why SeqArray anchors UK Biobank, All of Us, and TOPMed rare-variant workflows. The biobank-scale advantage over VCF is filtering and aggregating without re-streaming.

Parameters worth knowing. - seqSetFilter() — set sample / variant filters on the open GDS handle. - seqGetData() — extract data ("genotype", "annotation/info/AF", etc.) under the current filter. - seqAlleleFreq() — compute per-variant allele frequencies. - seqApply() / seqBlockApply() — stream a function over variants/samples in blocks.

Canonical example. The vignette's typical access pattern: 

gds <- seqOpen("mydata.gds")
seqSetFilter(gds, sample.id = samples, variant.id = variants)
genotypes <- seqGetData(gds, "genotype")
af        <- seqAlleleFreq(gds)
seqClose(gds)
The original Zheng et al. 2017 Bioinformatics paper benchmarks this against VCF on population-scale WGS.

Where it fits in the corpus

  • AACR 2026: axis = bioinfo / AI methods; biobank rare-variant infrastructure is upstream of any large-scale germline-cancer or pharmacogenomics analysis
  • EuroBioC 2025: no direct counterpart; biobank-scale GDS work is GBCC-side
  • ISMB 2026: scaffold; expect overlap on biobank methods
  • Nextflow Summit 2026: not mentioned

Notes

SeqArray anchors a multi-package ecosystem (SeqVarTools, SNPRelate, GENESIS) used across UK Biobank, All of Us, and TOPMed analyses. The Day-2 talk is an annotation-pipeline update layered on a mature substrate — the news is workflow ergonomics, not a new container.