henhaobio
Sign in

bioRxivBioinformaticsPreprint

Logan: Planetary-Scale Genome Assembly Surveys Life's Diversity

Rayan ChikhiiD, Téo LemaneiD, Raphaël Loll‐KrippleberiD, Mercè Montoliu-Nerin, Brice Raffestin, Antonio Pedro Camargo, Carson J. MilleriD, Mateus Bernabe FiamenghiiD, Daniel Paiva AgustinhoiD, Sina MajidianiD, Greg Autric, Maxime Hugues, Junkyoung Lee, Roland FaureiD, Kristen D. CurryiD, Jorge A. Moura de SousaiD, Eduardo P. C. RochaiD, David KoslickiiD, Paul MedvedeviD, Purav Gupta, Jessica SheniD, Alejandro Morales-Tapia, Kate Sihuta, Peter J. Roy, Grant W. BrowniD, , Artem BabaianiD

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

Abstract

The breadth of lifes diversity is unfathomable, but public nucleic acid sequencing data offers a window into the dispersion and evolution of genetic diversity across Earth. However the rapid growth and accumulation of sequence data have outpaced efficient analysis capabilities. The largest collection of freely available sequencing data is the Sequence Read Archive (SRA), comprising 27.3 million datasets or 5 x 1016 basepairs. To realize the potential of the SRA, we constructed Logan, a massive sequence assembly transforming short reads into long contigs and compressing the data over 100-fold, enabling highly efficient petabase-scale analysis. We created Logan-Search, a k-mer index of Logan for free planetary-scale sequence search, returning matches in minutes. We used Logan contigs to identify >200 million plastic-degrading enzyme homologs, and validate novel enzymes with catalytic activities exceeding current reference standards. Further, we vastly expand the known diversity of proteins (30-fold over UniRef50), plasmids (22-fold over PLSDB), P4 satellites (4.5-fold), and the recently described Obelisk RNA elements (3.7-fold). Logan also enables ecological and biomedical data mining, such as global tracking of antimicrobial resistance genes and the characterization of viral reactivation across millions of human BioSamples. By transforming the SRA, Logan democratizes access to the worlds public genetic data and opens frontiers in biotechnology, molecular ecology, and global health.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (1 Sept 2025) · CC BY 4.0

Assembling the Sequence Read Archive (27.3 million accessions) with over 100-fold compression enables minutes-scale k-mer search and mining of enzymes, plasmids, proteins and viral reactivation.

The largest public assembled collection, GenBank WGS, spans less than 3% of the Sequence Read Archive (SRA), and k-mer indexes of reads have not scaled beyond 9% of it (Table 1). Logan assembles 96% of the SRA and adds a k-mer search index over 23.4 million accessions.

Key results

  1. ShownNearly all of the SRA (96% by size) was assembled in the cloud into near-lossless unitigs and more contiguous contigs, shrinking the data more than 100-fold and giving 36 times more assembled sequence than GenBank WGS.

    44.1 petabases of reads became 0.9 petabases of contigs (4.59 Pbp of unitigs), in about 30 hours of wall-clock time at a peak of 2.18 million CPU cores; around 30 million CPU hours in total. · 1 Main; A Petabyte-Scale Assembly and Search Engine of…; 3.1.4 Assembly results

  2. ShownLogan-Search, a k-mer (k=31) index of Logan unitigs, finds which SRA accessions contain a query sequence of up to 1 kb within minutes.

    Index of 23.4M accessions, about 1 petabyte; a 1 kb query takes about 6 minutes on 12 small 4-vCPU virtual machines; Bloom filter false positive rate tuned to about 0.005%. · A Petabyte-Scale Assembly and Search Engine of…; 3.3 SRA-wide public search engine

  3. ShownA protein homology search from 213 known plastic-active enzymes yielded the PETadex collection of candidate enzymes, and a yeast activity screen found many active candidates, two of which outperformed controls.

    735 million Logan sequences (65.7%) not matching nr at 90% identity clustered into 215.7 million homologs (205-fold over nr). BHET screen: 35/161 (22%) natural and 8/21 (38%) ancestral sequences active; the top two exceeded IsPETase and gave 4-fold more TPA than FAST-PETase. · Expanding the arsenal of plastic-active enzymes; 3.5 Plastic active enzyme homolog expansion

  4. ShownSearching two Human Herpesvirus-6 (HHV-6) transcripts across human RNA-seq data revealed recurrent viral reactivation in tumor-infiltrating lymphocyte (TIL) therapy products, in addition to CAR T cell and organoid samples.

    1,476,236 human RNA-seq datasets queried; 13 BioProjects had >50% k-mer coverage (4 known positives, 9 new). High-confidence HHV-6 detection in 5 infusion products from 16 TIL donors profiled by RNA-seq; ChIP-seq of 19 donors showed reactivation in all three trials; 72 variants separated two donors' viral strains. · Characterization of Human Herpes Virus 6…; 3.12 Discovery and characterization of HHV-6…

  5. ShownMining Logan contigs expanded known proteins, plasmids, P4 satellites and Obelisk elements by several-fold, up to roughly 30-fold over reference sets.

    109.4 billion complete non-human proteins reduced to about 3 billion representatives (Logan50, 50% identity per Methods), nearly 30-fold over UniRef50; Obelisks up 3.7-fold to 26,263 sequences; P4 satellites 4.5-fold; plasmid phylogenetic diversity up to 21.8-fold vs PLSDB and 7.0-fold vs IMG/PR. · 1 Main; Billions of diverse Logan proteins; Obelisks; 3.13 Gene calling and protein clustering

Methods, data and reproducibility

Each SRA accession was assembled independently in containers on cloud batch computing (unitigs with a modified Cuttlefish2, contigs with Minia3). Contigs were searched with DIAMOND2 or minimap2, and unitigs were indexed with kmtricks/kmindex for k-mer queries. Selected enzyme candidates were tested in a yeast assay.

  • SRA accessions with public consent and read length of at least 31 bp, snapshot of 2023-12-10 · multiple species (SRA spans all known life) · n = 27,764,168 accessions
  • PAZy database of experimentally validated plastic-active enzymes used as search seed · n = 213 sequences
  • Candidate plastic-active enzymes from Logan, expressed in yeast for a BHET activity screen · Saccharomyces cerevisiae (expression host) · n = 161 natural, 21 ancestral reconstructions, 13 resampled
  • Human RNA-seq datasets searched for HHV-6 transcripts; TIL cultures profiled by RNA-seq and ChIP-seq · human · n = 1,476,236 RNA-seq datasets; 16 TIL donors (RNA-seq), 19 (ChIP-seq) · PRJNA901910 (RNA-seq), PRJNA901909 (ChIP-seq)
  • Viral proteins with poor default multiple sequence alignments, used to test Logan50 for structure prediction · virus · n = 100 proteins
  • Circular contigs screened for plasmids · bacteria and archaea (metagenomic and isolate samples) · n = 3,885,511 circular contigs, 252,507 samples; 468,614 putative plasmids
  • Initial Obelisk clusters used as queries for the Logan expansion · n = 1,744 clusters (7,202 circular genomes)
Methods & tools
Cuttlefish2, Minia3, KMC3, DIAMOND2, minimap2, kmindex, kmtricks, MMseqs2 Linclust, AlphaFold2/ColabFold, geNomad, Prodigal, HMMER
Data & accessions
https://registry.opendata.aws/pasteur-logan/s3://logan-pub/https://github.com/ababaian/petadexs3://logan-pub/paper/AMRs3://logan-pub/paper/plasmidss3://logan-pub/paper/Obelisk
Software & environment
Tool versions given in places (DIAMOND2 v2.1.9, minimap2 v2.28, kmindex v0.5.3, Prodigal 2.6.3); assembly runs in a Docker container on AWS Batch.
Assay conditions & compute
About 30 million CPU hours for assembly, peaking at 2.18 million cores. Each search across all Logan contigs takes about 11 hours on 60,000 vCPUs. The search index took about 10 days on 625 Azure VMs.
Reproducibility gaps
The authors say the assembly cloud infrastructure is intended for internal use because of its high execution cost, although its source code is public.
Limitations (5)
  • Some accessions produced too many unitigs to fit the assembly graph in memory and were not assembled into contigs: 26.8 million accessions (88% of SRA size) have contigs versus 27.3 million (96%) with unitigs. · stated by the authors
  • Unitig abundances are approximate, relying on heuristics (k+1-mer summing and 8-bit encoding), and k-mers seen only once per accession were discarded. · stated by the authors
  • Logan-Search natively returns matching accessions rather than matching sequences (a separate back-to-sequences microservice retrieves those), and its Bloom filters have a non-zero false positive rate (about 0.005%). · stated by the authors
  • Enzyme activity was tested only on the PET subunit BHET in yeast with a small number of tested candidates, so activity on PET plastic itself is not shown. · noted in this summary
  • HHV-6 reactivation in TIL products is inferred from sequencing data of a limited number of donors per trial, without independent experimental confirmation in the text. · noted in this summary

Read this if you want to search or mine all public sequencing data for genes, enzymes, plasmids or viruses without assembling reads yourself.

Citations
76

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
THISSteinegger '19Bolger '17Vaser '19Desvillecha '26Chin '19Karasikov '25
55% semantic matchBioinformatics
Martin Steinegger, Milot Mirdita, Johannes Söding·Nature Methods·Jun 24, 2019· 445 citations