Unicycler: resolving bacterial genome assemblies from short and long sequencing reads
Ryan R. WickiD, Louise M. JuddiD, Claire L. GorrieiD, Kathryn E. HoltiD
Abstract
1.The Illumina DNA sequencing platform generates accurate but short reads, which can be used to produce accurate but fragmented genome assemblies. Pacific Biosciences and Oxford Nanopore Technologies DNA sequencing platforms generate long reads that can produce more complete genome assemblies, but the sequencing is more expensive and error prone. There is significant interest in combining data from these complementary sequencing technologies to generate more accurate \"hybrid\" assemblies. However, few tools exist that truly leverage the benefits of both types of data, namely the accuracy of short reads and the structural resolving power of long reads. Here we present Unicycler, a new tool for assembling bacterial genomes from a combination of short and long reads, which produces assemblies that are accurate, complete and cost-effective. Unicycler builds an initial assembly graph from short reads using the de novo assembler SPAdes and then simplifies the graph using information from short and long reads. Unicycler utilises a novel semi-global aligner, which is used to align long reads to the assembly graph. Tests on both synthetic and real reads show Unicycler can assemble larger contigs with fewer misassemblies than other hybrid assemblers, even when long read depth and accuracy are low. Unicycler is open source (GPLv3) and available at github.com/rrwick/Unicycler.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 8,731
- FWCI
- 138.36
- Field percentile
- top 1%
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · max effort · 8 Nov 2026 · from the full text of version 1 (22 Dec 2016) · CC BY 4.0
Unicycler scaffolds SPAdes short-read assembly graphs using long reads, producing more complete bacterial assemblies with fewer misassemblies than competing hybrid tools, even at low long-read depth.
Illumina-only assemblies of bacteria fragment at repeats longer than short-read fragments, long-read-only sequencing is costly and error-prone, and existing scaffolders that join contigs with paired or long reads often introduce misassemblies. Unicycler instead resolves repeats on the short-read assembly graph itself, so bridges keep short-read base accuracy and low long-read depth can complete genomes of isolates already sequenced on Illumina.
Key results
ShownOn simulated hybrid read sets, Unicycler had the fewest misassemblies and the highest mean NGA50 of the compared assemblers at every long-read depth, often reaching complete or near-complete genomes from sparse long-read coverage.
Hybrid tests (2520 per assembler, 12 reference genomes, five replicates): npScarf and Cerulean consistently had 10 or more misassemblies and SPAdes erred on Shigella genomes rich in ~1 kbp insertion-sequence repeats; Unicycler misassemblies were lowest, scaling with mode, and it was often near-complete at 4x long-read depth. · 4.2. Simulated read sets; 4.4. Hybrid assembly of simulated long and short…; Figure legends
ShownReal E. coli K-12 MG1655 reads reproduced the simulated trends: Unicycler reached more complete assemblies at lower long-read depth, and its contiguity depended more on long-read length than on per-base accuracy.
Six long-read sets, 20 random subsamples per depth: best NGA50 2.0 Mbp for Unicycler vs 1.4 Mbp for SPAdes; Unicycler did worst on PacBio RS (shortest reads) rather than ONT R7 (lowest identity) and best on PacBio RS II C3; in 8x simulations read length significantly shifted NGA50 (two-tailed t-test) while accuracy had a weaker effect (one-way ANOVA). · 4.4. Hybrid assembly of simulated long and short…; 4.6. Real E. coli K-12 read sets; Figure legends
ShownTracking ONT R9 sequencing of K. pneumoniae INF125 minute by minute, Unicycler reached a complete three-contig assembly (chromosome plus two plasmids) sooner and at lower depth than npScarf, hybridSPAdes or miniasm.
240 cumulative per-minute subsets of a four-hour run (156 Mbp, 28.6x, read N50 11,470 bp): Unicycler finished at 45 minutes (5.3x) vs 76, 102 and 213 minutes for npScarf (9.0x), SPAdes (12.1x) and miniasm (25.3x); SPAdes added erroneous contigs, and Unicycler and SPAdes were modestly more accurate than npScarf by Illumina read alignment. · 4.8. Klebsiella pneumoniae ONT assembly case…; Figure legends
ShownWith short reads alone, Unicycler beat SPAdes and ABySS on each QUAST metric; the authors credit its wide k-mer sweep for contiguity and its selective use of SPAdes repeat resolution for fewer misassemblies.
360 assemblies per assembler: Unicycler typically chose k = 95 vs the automatic 21 to 55 range in SPAdes and k = 64 for ABySS, giving the highest NGA50 in normal and bold modes; pre-repeat-resolution SPAdes graphs had very low misassembly rates, implicating repeat resolution as the main error source. · 4.3. Assembly of simulated short read datasets; Figure legends
ShownFor multidrug-resistant K. pneumoniae INF274 (Illumina plus PacBio RS II), only Unicycler, versus hybrid SPAdes, HGAP and Canu, output circularized replicons in graph form without duplicated start/end sequence.
Only plasmids 5 and 6, which share sequence and had few PacBio reads, stayed incomplete in Unicycler; SPAdes had the same problem and also failed on plasmid 3 (prophage shared with the chromosome); SPAdes showed slight and HGAP and Canu significant end overlaps (read-depth drops), and Canu circularized no replicons. · 4.7. Klebsiella pneumoniae PacBio assembly case…; Figure legends
Methods, data and reproducibility
Unicycler picks the best SPAdes graph from a 10-value k-mer sweep, infers contig multiplicity from depth and connectivity, bridges single-copy contigs via paired-end repeat resolution and semi-global long-read alignment, then rotates circular replicons to dnaA/repA and Pilon-polishes. QUAST benchmarks compared it with SPAdes, ABySS, npScarf and Cerulean.
- Simulated HiSeq 2500-like paired-end reads (ART; 125 bp, 400 bp inserts) and PBSIM long reads (60%, 75%, 90% accuracy; 10 or 25 kbp; 0.25x to 16.0x depth) · seven bacterial species and Saccharomyces cerevisiae · n = 12 reference genomes; five replicates; 16920 assemblies
- Real E. coli K-12 substr. MG1655 Illumina MiSeq reads combined with long reads from ONT R7, ONT R9 (pass and fail), PacBio RS, and PacBio RS II C2 and C3 chemistries · Escherichia coli K-12 substr. MG1655 · n = six long-read sets; 20 subsampled trials per depth
- Sanger-sequenced E. coli K-12 substr. MG1655 reference genome (University of Wisconsin, 1997) used as QUAST ground truth · Escherichia coli K-12 substr. MG1655 · NC_000913.3
- Clinical urine isolate INF274 (multidrug-resistant ST340, CG258): Illumina HiSeq 2500 125 bp paired-end reads (750 Mbp) and PacBio RS II long reads (1.3 Gbp) · Klebsiella pneumoniae · n = one isolate
- Clinical urine isolate INF125 (virulent ST45): ONT R9 reads from a four-hour run (156 Mbp, 28.6x, N50 11,470 bp) split into 240 cumulative subsets, plus Illumina reads · Klebsiella pneumoniae · n = one isolate; 240 read subsets
- Methods & tools
- Unicycler (incl. unicycler_align; unicycler_polish using Pilon, GenomicConsensus, FreeBayes, ALE), SPAdes (v3.6.2 or later inside Unicycler; v3.8.1 hybrid mode as comparator), SeqAn C++ library (semi-global alignment, consensus), Bowtie2 and Pilon (short-read polishing), TBLASTN (dnaA/repA search for replicon rotation), ABySS v1.9.0, npScarf v1.6-01c, Cerulean v0.1, HGAP v3 and Canu v1.3 (PacBio-only assemblers), miniasm (long-read-only assembler), QUAST v4.3 (strict-NA mode), ART v2.5.8 and PBSIM (read simulation)
- Software & environment
- Built on SPAdes (v3.6.2 or later), SeqAn, TBLASTN, Bowtie2 and Pilon; GPLv3; tests used SPAdes v3.8.1, ABySS v1.9.0, npScarf v1.6-01c, Cerulean v0.1, HGAP v3, Canu v1.3, QUAST v4.3, ART v2.5.8; ONT pass/fail via Metrichor v0.16.37960.
- Assay conditions & compute
- All assembly tests ran on 8 CPU cores with 16 GB RAM; median Unicycler runtime 46 minutes for 8x synthetic hybrid tests vs 8 minutes for SPAdes and npScarf and 23 minutes for Cerulean; Unicycler maximum 13 hours (A. baumannii A1); runs capped at 24 hours.
- Reproducibility gaps
- No accession numbers or download links appear in the text for the newly sequenced K. pneumoniae INF274 and INF125 reads or the simulated read sets; real E. coli read sets are described only in Table 2 and exact commands in Table S1.
Limitations (6)
- Performance hinges on high-quality short reads: unsequenced regions leave dead ends in the assembly graph, and bridges that must fall back on long-read consensus sequence across such gaps are more error-prone. · stated by the authors
- Slower than competing hybrid assemblers (median 46 minutes vs 8 minutes for SPAdes and npScarf on 8x synthetic tests; up to 13 hours for A. baumannii A1) and not yet capable of human-scale genome or metagenome assembly. · stated by the authors
- Replicons that share sequence and lack long-read coverage cannot be scaffolded apart, as with INF274 plasmids 5 and 6, which stayed incomplete in both Unicycler and hybrid SPAdes. · stated by the authors
- Real E. coli tests are prone to false-positive misassembly calls because sequenced subcultures genuinely differ from the 1997 reference, likely depressing NGA50 (best 2.0 Mbp) relative to simulated reads. · stated by the authors
- Most quantitative comparisons rest on ART/PBSIM simulations with fixed error settings; real-data tests cover a single E. coli strain scored against an imperfect reference and K. pneumoniae isolates evaluated without a ground-truth genome. · noted in this summary
- The ONT time-to-completion advantage (45 vs 76 minutes for npScarf) comes from a single isolate and sequencing run, so its robustness to other read-length distributions, genomes and plasmid contents is untested. · noted in this summary
Who should read:Read this if you want to complete bacterial genomes by adding low-depth PacBio or ONT reads to existing Illumina data, or you build graph-based hybrid assemblers or long-read-to-graph aligners.
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
- Comparison of long-read sequencing technologies in the hybrid assembly of complex bacterial genomes
Nicola De Maio, Liam P. Shaw, … Nicole E Stoesser·Microbial GenomicsPeer-reviewed·Sep 1, 2019
- Do You Want to Build a Genome? Benchmarking Hybrid Bacterial Genome Assembly Methods
Georgia L Breckell, Olin Silander·bioRxiv·Nov 7, 2021
- Trycycler: consensus long-read assemblies for bacterial genomes
Ryan R. Wick, Louise M. Judd, … Kathryn E. Holt·Genome biologyPeer-reviewed·Sep 14, 2021
- Autocycler: long-read consensus assembly for bacterial genomes
Ryan R. Wick, Benjamin P. Howden, Timothy Paul Stinear·BioinformaticsPeer-reviewed·Aug 25, 2025
- Technical report on best practices for hybrid and long read de novo assembly of bacterial genomes utilizing Illumina and Oxford Nanopore Technologies reads
Simon T. Hackl, Theresa Anisja Harbig, Kay Nieselt·bioRxiv·Oct 25, 2022
- Efficiently Constructing Complete Genomes with CycloneSEQ to Fill Gaps in Bacterial Draft Assemblies
Hewei Liang, Yuanqiang Zou, … Liang Xiao·GigabytePeer-reviewed·Apr 25, 2025