henhaobio
Sign in

The ISME JournalBioinformaticsPeer-reviewed journal article

dRep: A tool for fast and accurate genome de-replication that enables tracking of microbial genotypes and improved genome recovery from metagenomes

Matthew R. OlmiD, Christopher T. BrowniD, Brandon BrooksiD, Jillian F. BanfieldiD

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

Abstract

The number of microbial genomes sequenced each year is expanding rapidly, in part due to genome-resolved metagenomic studies that routinely recover hundreds of draft-quality genomes. Rapid algorithms have been developed to comprehensively compare large genome sets, but they are not accurate with draft-quality genomes. Here we present dRep, a program that sequentially applies a fast, inaccurate estimation of genome distance and a slow but accurate measure of average nucleotide identity to reduce the computational time for pair-wise genome set comparisons by orders of magnitude. We demonstrate its use in a study where we separately assembled each metagenome from time series datasets. Groups of essentially identical genomes were identified with dRep, and the best genome from each set was selected. This resulted in recovery of significantly more and higher-quality genomes compared to the set recovered using the typical co-assembly method. Documentation is available at http://drep.readthedocs.io/en/master/ and source code is available at https://github.com/MrOlm/drep.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (13 Feb 2017) · CC BY 4.0

dRep groups genomes with fast Mash distances, then accurate ANI, de-replicating 1,125 infant gut genomes in 92 CPU hours versus 2,784 for naive gANI.

Accurate average nucleotide identity (ANI) tools such as gANI become very slow when comparing every pair in large sets of draft genomes from metagenomes, and fast Mash estimates lose accuracy as bin completeness falls. dRep makes de-replication practical, so each sample can be assembled on its own and the results then de-replicated, which avoids the fragmentation that co-assembly causes when related strains co-occur.

Key results

  1. ShownCoarse Mash clustering followed by pairwise ANI only within each primary cluster sharply cut de-replication compute versus naive gANI; simulation predicted gains of multiple orders of magnitude for larger sets, and the measured runtime closely matched the simulation.

    1,125 genomes from 195 fecal metagenomes (21 premature infants): 92 CPU hours observed vs 93 CPU hours predicted by simulation, compared with 2,784 hours for naive gANI. · Abstract; Figure legends

  2. ShownMash ANI estimates are unreliable for incomplete genomes. ANIm and gANI stay accurate once genomes overlap by about half or more, which is why dRep uses Mash only for the first clustering step.

    A complete Escherichia coli genome was subset ten times in 10% increments and the subsets compared pairwise (100 total comparisons) with ANIm, MASH and gANI. ANIm and gANI were accurate at ≥50% overlap; MASH was accurate only for essentially complete genomes. · Abstract; Figure legends

  3. ShownAssembling each sample separately, adding co-assembly bins and de-replicating with dRep gave more and better genome bins than co-assembly alone.

    34% more bins (≥50% complete, ≤25% contaminated) than co-assembly alone. For genomes recovered both ways, individual assemblies were less fragmented (N50, p = 9.6e-13) and more complete (p = 2.3e-8), by Wilcoxon signed-rank test. · Abstract; Figure legends

  4. ShownPairwise ANI comparison in dRep found bacterial strains shared across different preterm infants, so the tool can be used for strain tracking.

    Strains were defined as the same when at least 50% of the genome aligned at 99.9% ANI. 10 strains were found in at least 3 of the 21 infants. · Abstract; Figure legends

  5. SuggestedIn one infant's time series with known strain heterogeneity, co-assembly gave a contaminated Staphylococcus hominis bin and a more fragmented S. aureus genome. The gaps fell in regions with population-level SNP variation.

    S. hominis: the co-assembly bin had duplicate and triplicate copies among 104 checkM single-copy genes. Earlier anvi'o curation reached 73% complete, 6.6% redundant, vs 98% complete, 0% redundant for the uncurated individual-assembly bin. S. aureus: co-assembly scaffolds were more fragmented against a 2.7 Mbp reference. · Abstract; Figure legends

Methods, data and reproducibility

Genomes are filtered by checkM completeness, then split into primary clusters by Mash distance (90% Mash ANI cutoff). Within each primary cluster, genomes are compared pairwise with gANI or ANIm and grouped into secondary clusters (99% ANI) by SciPy hierarchical clustering, and the highest-quality genome in each cluster is kept.

  • Genome bins from infant fecal metagenomes, assembled per sample and by co-assembly, used to measure runtime, genome recovery and strain tracking · human (premature infant gut microbiome) · n = 1,125 genomes from 195 metagenomes of 21 infants
  • Single-infant metagenome time series with known strain heterogeneity, used to compare co-assembly with individual assembly · human infant gut microbiome (Staphylococcus hominis, S. aureus bins)
  • Complete genome subset in 10% increments to test how accurate ANIm, MASH and gANI are on partial genomes · Escherichia coli · n = 100 pairwise comparisons
  • Complete reference genome (2.7 Mbp) used to align scaffolds from co-assembled and individually assembled bins · Staphylococcus aureus
Methods & tools
dRep (version 0.3.3), Mash (MinHash distance), gANI, ANIm, MUMmer, CheckM, SciPy hierarchical clustering, anvi'o (prior manual bin curation), Wilcoxon signed-rank test
Reagents & environment
dRep version 0.3.3 is supplied as a supplemental clone. It depends on checkM, Mash, gANI or ANIm (MUMmer) and SciPy.
Assay conditions & compute
The infant gut set took 92 CPU hours (93 predicted by simulation) vs 2,784 hours for naive gANI. Hardware is not specified.
Reproducibility gaps
No accession numbers or download links are given for the 195 infant metagenomes or the recovered genome bins, and the hardware used for runtime measurements is not stated.
Limitations (4)
  • The first step relies on Mash, so dRep can only be used when the genomes being compared are more than 50% complete. · stated by the authors
  • checkM has no established completeness metrics for viruses and plasmids, so users must estimate their completeness with a separate method. · stated by the authors
  • The 34% gain compares a pooled set (individual-assembly plus co-assembly bins) with co-assembly alone, so it does not isolate the effect of individual assembly by itself. · noted in this summary
  • Testing used only low-diversity preterm infant gut data, and the strain-heterogeneity effects rest on one infant and two Staphylococcus species. The larger speedups the authors expect in more diverse environments were not tested. · noted in this summary

Read this if you assemble metagenomes sample by sample and need to collapse redundant draft genomes, choose the best representative of each, or track strains shared across hosts.

Citations
3,024
FWCI
22.89
Field percentile
top 1%

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
THISYazhini '25Evans '20Sieber '17Salamzade '25Viehweger '19Bushnell '26
70% semantic matchBioinformatics
Arangasamy Yazhini, Johannes Söding·Bioinformatics·Sep 26, 2025· 2 citations