Communication Dans Un Congrès Année : 2026

Phylogenetic ordering and batching of million-genome collections for reduced storage and faster retrieval

Résumé

Modern bacterial genome collections such as AllTheBacteria [1] contain millions of genomes and continue to grow rapidly. Efficient compression of such collections critically depends on genome ordering to co-localize shared sequence redundancy prior to low-level compression, with phylogenetic compression representing the current state of the art [2]. However, MiniPhy, the approach used for AllTheBacteria, performs phylogenetic reordering only within fixed-size batches, while batching itself relies on ad hoc heuristics based on metadata such as accession order. This strategy degrades for highly sampled species, where closely related genomes are no longer sufficiently accession-local, and is not suitable for applications that require explicit control over batch size, such as random genome access. Here, we present Phylopack, a scalable method for phylogenetic reordering that improves both compression efficiency and random-access genome retrieval. We formalize phylogenetic reordering as an optimization problem and introduce an efficient approximation that combines instance-size-constrained phylogenetic inference with sketch-based nearest-neighbor identification to reconstruct approximate species-level phylogenies at scale. Applied to AllTheBacteria, Phylopack reduces the compressed size of major species (93.5% of the collection) to 15 GB with MBGC [3] – a four-fold improvement over compression with current MiniPhy batching. When applied to random genome access, Phylopack achieves an 18× improvement in AGC[4]/XZ single-genome retrieval compared to default compression (≈0.5 s/genome), with negligible impact on collection size. Overall, Phylopack presents a substantial advance in phylogenetic compression, extending its practical applicability to arbitrary million-genome bacterial collections while enabling near-real-time genome retrieval.

Fichier principal
Vignette du fichier
DSB-2026-PhyloPack.pdf (56.92 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05584462 , version 1 (08-04-2026)

Licence

Identifiants

  • HAL Id : hal-05584462 , version 1

Citer

Tam Truong, Dominique Lavenier, Pierre Peterlongo, Karel Břinda. Phylogenetic ordering and batching of million-genome collections for reduced storage and faster retrieval. DSB 2026 - workshop Data Structures in Bioinformatics, Feb 2026, Venice, Italy. ⟨hal-05584462⟩
37 Consultations
16 Téléchargements

Partager

  • More