Skip to content
hub Computational Genomics

Graphs over
genomes.

Long-read sequencing, pangenome graphs, and the structural variation hidden between reference assemblies.

terminal View software article Recent papers
$ cargo install graphmap2
// fig.01   long-read assembly → graph construction → variant projection
// Research overview

Research Overview

// Program 01 — Pangenome graphs

Our core program develops methods for pangenome graph construction and querying at the scale of hundreds to thousands of assemblies. We work with long-read data (Oxford Nanopore R10 and PacBio HiFi) because the structural-variant signal is in reads that short-read pipelines never see. The challenge is no longer assembly — it is the comparison.

// Program 02 — Non-model SV detection

A second line of work targets structural-variant detection in non-model organisms — pollinators, fungi, livestock — where the reference is either fragmentary or wrong in ways that bias every downstream call. We build graph-based callers that treat the reference as one path among many and report variation against the graph rather than against a single haplotype.

// Program 03 — Clinical pangenomes

We also collaborate on clinical pangenomes with partner hospitals, focused on diagnostically relevant medically complex loci (HLA, KIR, the MHC region) where a single linear reference is known to mislead. All methods we publish ship as open-source tooling — see the software page.

// Releases

What we ship

See all software arrow_forward
// Latest paper 2026 Nature Methods

pansift: a succinct index for population-scale pangenome graphs of vertebrate genomes

Mateusz Kowalski†, Søren Lindqvist, Ezekiel Holt*

Holding a vertebrate-scale pangenome graph in memory across thousands of haplotype paths has been the bottleneck for graph-coordinate variant work. pansift is a succinct index that fits the full graph and path set in roughly 18 bytes per node, with sub-millisecond path queries. The hard part was preserving random-access lookup over arbitrary path subsets without rebuilding — we describe a layered rank/select scheme that trades a small construction-time cost for query latency that does not degrade with cohort size.

code_blocks Repository link DOI
// Activity

Recent news

  1. package_2 release 2026-04-22

    svhound 0.7 released — adds support for long-read multi-sample joint calling.

  2. article paper 2026-04-05

    New paper in Nature Methods on the joint pangenome–variant calling framework.

  3. description preprint 2026-03-12

    Preprint on honeybee pangenome assembly available on bioRxiv.

  4. mic talk 2026-02-18

    Ezekiel speaks at a biological data science meeting in November 2026.

  5. person_add hire 2026-01-10

    Welcome Yusuf Aydın, joining as a Research Software Engineer.

// Open positions

Join the lab

We hire on a rolling basis. We are particularly interested in candidates comfortable with low-level systems work AND population genomics — the intersection is small but it is where our problems live.

  • PhD Candidate — Pangenome graph algorithms

    4 years · structured PhD programme · starts Sep 2026

    Apply by 31 January 2026.

  • Postdoctoral Researcher — Structural variant population genetics

    3 years · externally funded

    Rolling.

  • Research Software Engineer — Reference implementations

    Permanent · institute staff

    Open until filled.

mail Contact the lab
// Follow the lab