henhaobio
Sign in

bioRxivBioinformaticsPreprint

AllTheBacteria: a community resource empowers biology and discovers novel peptide antibiotics

Martin HuntiD, Marcelo D. T. TorresiD, Nabil-Fareed AlikhaniD, Daniel AndersoniD, Maria Luiza AndreaniiD, Josefin BlomiD, George BourasiD, Fiona S.L. BrinkmaniD, Laura M. CarrolliD, Matthew A. CroxeniD, R. Andres FlotoiD, Michael B. HalliD, Jane HawkeyiD, Samuel HorsfieldiD, Baofeng JiaiD, Jake A. LaceyiD, Hyun-Su Lee, Leandro LimaiD, Neil MacAlasdairiD, Sudaraka MallawaarachchiiD, William MatlockiD, Ahmed M. MoustafaiD, Robert Petit, Vignesh RamnathiD, Vishnu RaghuramiD, , Zamin IqbaliD

PDFDOICodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Public microbial genomes encode an immense record of biological diversity, evolution and molecular function, but much of this information remains difficult to reuse because raw sequencing data are not uniformly assembled, quality controlled, annotated or searchable at scale. Here we present AllTheBacteria, an open, community-built resource that transforms public bacterial short-read whole-genome sequencing reads into a uniformly processed discovery platform. The current analysed release contains 2,440,377 high-quality bacterial and archaeal genomes from 11,273 species, together with standardized taxonomic assignments, genome annotations, antimicrobial resistance calls, antiphage-defence annotations, protein structure predictions and AI-ready sequence tables. We show that this infrastructure enables applications that would otherwise be impractical, from global sequence search and outbreak contextualization to pangenome method development, antimicrobial resistance reservoir mapping and antiphage-defence ecology. As a stringent experimental demonstration, we mined 3,919,096 encrypted peptide fragments from AllTheBacteria proteomes using our deep learning model APEX 1.1, identifying 1,867 candidates with predicted antimicrobial activity. We synthesized 24 representative peptides and tested them against 20 clinically relevant bacterial strains, including antibiotic-resistant pathogens. Multiple peptides showed low-micromolar activity, membrane-responsive conformational transitions and selective envelope perturbation. A lead molecule, ATB20, reduced Acinetobacter baumannii burden in a murine skin abscess model with efficacy comparable to polymyxin B and no overt toxicity. Together, these results establish AllTheBacteria as both a foundational community resource for microbiology and a renewable engine for AI-guided antimicrobial discovery.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 8 (7 Jul 2026) · CC BY 4.0

A uniformly processed set of 2,440,377 bacterial and archaeal genomes with search tools and annotations, plus an AI-mined peptide set yielding ATB20, active in a mouse skin infection model.

Public bacterial reads in INSDC are not consistently turned into harmonised assemblies, quality metrics, annotations or searchable products, so many operations stay limited to groups with large compute. AllTheBacteria provides an open, community-governed, uniformly processed atlas with search indexes and AI-ready tables.

Key results

  1. ShownThe release covers 2,440,377 genomes from 11,273 species, with companion tools (MiniPhy compression, LexicMap search, sketchlib.rust neighbour search) that make the collection distributable and searchable on modest hardware.

    MiniPhy stores 2.4 million genomes in 114 GB, versus 4 terabytes if each were gzipped individually; LexicMap searches in seconds to minutes with ~10 GB RAM; 100 near neighbours of a 2026 UK meningococcal outbreak genome were found in seven seconds. · A community-built atlas of bacterial and…; Data volumes inspire new algorithms and AI-ready…; MiniPhy; sketchlib.rust

  2. ShownResistance burden in non-priority species varies strongly by genus. Most rare genera carried one to two AMR genes per genome, but a few non-ESKAPEE genera (Mannheimia, Citrobacter, Bacillus, Vibrio) had higher medians of 4 to 7, versus ESKAPEE medians of 5 to 15.

    1,087 non-priority species scanned; 17 of 27 rare genera carried one to two known AMR genes per genome; ESKAPEE genus medians ranged from 5 (Escherichia) to 15 (Klebsiella, Enterococcus); Mannheimia, Citrobacter, Bacillus and Vibrio had medians of 7, 5, 4 and 4. · AMR reservoirs beyond priority pathogens; Antimicrobial resistance gene and mutation…

  3. SuggestedAcross the full phylogeny, 90% of the significant SimPhyNI interactions were avoidances (the authors attribute this partly to the method's reliance on convergent signals), whereas within E. coli the 101 interactions were split almost equally between avoidance and co-occurrence.

    Antiphage-defence systems profiled in 1,166,956 genomes from 27 Enterobacterales species, 226 systems, 22 present in every species; SimPhyNI found 39 significant interactions between 15 systems overall. In S. enterica, 7.05 systems per genome vs 7.91 in RefSeq complete assemblies; prevalence Pearson R=0.90. · Antiphage defense element prevalence and…; Defence System detection; Defence system association and avoidance testing

  4. ShownMining proteome fragments with the APEX 1.1 deep learning model produced 1,867 predicted antimicrobial candidates; synthesized peptides were active, most strongly against Gram-negative strains including resistant isolates.

    3,919,096 fragments screened; 24 peptides tested against 20 strains. ATB5, ATB3 and ATB12 had MIC of 2 to 4 μmol/L against E. coli isolates including a colistin-resistant one; ATB5 inhibited A. baumannii ATCC 19606 at 1 μmol/L; ATB51 inhibited MRSA at 8 μmol/L. · AI mining of AllTheBacteria reveals encrypted…; APEX predictions; Minimal Inhibitory Concentration (MIC)…

  5. SuggestedThe lead peptide ATB20 lowered A. baumannii burden in a mouse skin abscess model to a degree comparable to polymyxin B, without overt weight loss.

    Single topical dose at 10×MIC, six mice per group: ~2-log lower burden at day 2 and ~1.5-log at day 4 vs untreated controls, matching polymyxin B; body weight stayed stable. · Anti-infective activity of an ATB peptide in…; Skin abscess infection mouse model; Quantification and statistical analysis

Methods, data and reproducibility

Public ENA paired-end Illumina isolate reads were assembled, species-assigned, quality-filtered and annotated with a common pipeline, then screened with tools for AMR, defence systems and other features. Proteome fragments were scored with APEX 1.1, and 24 selected peptides were tested in vitro and in a mouse model.

  • ENA paired Illumina bacterial and archaeal isolate whole-genome sequencing reads, processed into assemblies · bacteria and archaea · n = 2,440,377 genomes from 11,273 species
  • Prior 661k dataset of uniformly assembled ENA isolates (Blackwell et al. 2021), reprocessed and included · bacteria · n = 661,384 samples in AllTheBacteria originate from it
  • Enterobacterales genomes for defence-system detection (species with 400 or more genomes) · bacteria (Enterobacterales) · n = 1,166,956 genomes, 27 species
  • Hypothetical proteins with AlphaFold2 structure predictions · bacteria and archaea · n = 17,711,165 unique proteins
  • Peptide fragments from Bakta-annotated proteins scored by APEX 1.1 · bacteria and archaea · n = 3,919,096 fragments; 1,867 candidates; 24 tested
  • Antibacterial panel of clinically relevant strains, including resistant ones · bacteria · n = 20 strains
  • HEK293T cytotoxicity assay of the peptide panel · human (HEK293T cell line) · n = three biological replicates
  • Skin abscess infection model with A. baumannii ATCC 19606, female CD-1 mice · mouse · n = six mice per group
Methods & tools
Shovill, sylph, CheckM2, Bakta, AMRFinderPlus, DefenseFinder, SimPhyNI, MiniPhy, LexicMap, sketchlib.rust, ColabFold, APEX 1.1
Software & environment
Versions stated: Shovill v1.1.0, sylph 0.5.1 (GTDB r214), CheckM2 1.0.1, Bakta v1.9.4 (db 5.1), AMRFinderPlus v3.12.8, DefenseFinder v1.3.0, SimPhyNI v1.0.2, ColabFold v1.5.5, GraphPad Prism v.10.4.
Assay conditions & compute
sylph needed 10 Gb RAM and ~1 minute per sample; LexicMap ~10 GB RAM; structure prediction ran on AMD MI250x GPUs on Setonix (Pawsey); Bakta output of 35Tb was reduced to 1.5TB.
Reproducibility gaps
No code or model link for APEX 1.1 or the peptide analyses appears in the main text; the 24 peptide sequences are in a supplementary table.
Limitations (5)
  • The collection over-represents organisms that are sequenced frequently for research and surveillance, particularly human pathogens, so it does not sample bacterial diversity evenly. · stated by the authors
  • SimPhyNI relies on convergent signals, so across the full phylogeny it favours detecting avoidance over lineage-restricted co-occurrence, which affects interpretation of the interaction results. · stated by the authors
  • Some membrane-active ATB peptides had narrow therapeutic windows in HEK293T cells and may need chemical optimization to reduce toxicity to mammalian cells. · stated by the authors
  • Continued maintenance of the resource depends on sustained community effort and funding. · stated by the authors
  • In vivo validation covers one peptide, one A. baumannii strain, one topical dose and six mice per group; only 24 of 1,867 predicted candidates were tested, so the overall hit rate is not established. · noted in this summary

Read this if you need a uniformly processed, searchable set of public bacterial genomes, or want a worked example of mining proteomes for antimicrobial peptides.

Citations
77

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISTorres '23Santos-Juni '23Li '26Otani '25Veldsman '25Kang '25
72% semantic matchBioengineering
M. D. T. Torres, E. F. Brooks, A. Cesaro+4 more·bioRxiv·Aug 31, 2023