henhaobio
Sign in

bioRxivBioinformaticsPreprint

An efficient error correction and accurate assembly tool for noisy long reads

Jiang Yong HuiD, Zhuo WangiD, Zongyi SuniD, Benxia HuiD, Adeola Oluwakemi AyoolaiD, Fan LiangiD, Jingjing LiiD, José R. SandovaliD, D.N. CooperiD, Kai YeiD, Jue RuaniD, Chuan‐Le XiaoiD, De-Peng Wang, Dong‐Dong WuiD, Sheng Wang

PDFDOI

In-depth summary

Claude Opus 5.5 · xhigh effort · 17 Oct 2026 · from the full text of version 1 (12 Mar 2023) · CC BY-NC-ND 4.0

NextDenovo corrects noisy Nanopore reads faster and more accurately than Canu and Necat; its maize and 35 human-genome assemblies are more contiguous, with fewer misassemblies, than Flye's.

Oxford Nanopore (ONT) reads are cheap, long and suited to population-scale pan-genome projects but error-prone, and correction-then-assembly tools such as Canu and Necat tend to separate repeat copies better than assembly-then-correction tools like Flye and Wtdbg2 but run slower. NextDenovo is presented as a fast correction-first tool that adds iterative re-correction of low-score regions and repeat-aware graph cleaning, aimed at large repeat-rich genomes and population-scale human assembly.

Key results

  1. ShownRead-level correction runs faster than Canu and Necat and leaves fewer errors: corrected reads average above 99% accuracy, stay longer and carry fewer chimeric alignments.

    Human chr1, simulated / real ultra-long ONT: 7.44x / 69.25x faster than Canu (v2.0), 1.13x / 1.63x faster than Necat (v0.0.1); error rate 1.82% / 1.31% lower than Canu, 0.35% / 0.09% lower than Necat. Real data: 89.31% of reads >=97% accurate (Canu 80.64%, Necat 88.67%, raw 0.18%), mean length 90.98 kb (Canu 84.27, Necat 89.17). 89.10% of corrected reads map at >=99% coverage (Canu 85.13%, Necat 88.30%). · Benchmarking the error correction module; Evaluating error correction

  2. ShownOn the repeat-rich rice and maize genomes NextDenovo gave the most contiguous assemblies with the fewest misassemblies and higher QV; on the simpler Arabidopsis and fly genomes it performed on par with Necat and Flye.

    QUAST (v5.2.0) vs Necat, Canu, Flye, Wtdbg2: NGA50 about 2, 61, 15 and 758 times theirs (stated after the Z. mays comparison); LGA90 of 10 (A. thaliana) and 20 (O. sativa). A. thaliana: 2 more misassemblies and slightly lower QV than Flye. Z. mays run time: 23 times faster than Canu, slightly faster than Necat, slower than Flye; Wtdbg2 fastest on all genomes. · Assembly evaluation on non-human genomes; Evaluating assemblies

  3. ShownAcross 35 newly sequenced human ONT genomes, NextDenovo assemblies were more contiguous than Flye's, had fewer misassemblies and kept more multi-copy genes, at similar size and marginally higher base accuracy.

    Both ~2.83 Gb with ~90.84% genome coverage; single-copy genes 97.99% vs 97.82%, multi-copy genes 39.60% vs 33.93%; NGA50 1.03-1.61-fold larger and LGA50 68.18%-96.97% of Flye's in all 35 genomes; mean 388 misassemblies, ~70% of Flye's; QV 28.17 vs 28.06. Each genome: >150 Gb ONT reads (mean 21 kb), ~0.49x ultra-long (>100 kb) coverage, ~150 Gb short reads for polishing. · Assembly of 35 human genomes by NextDenovo and…; Evaluating assemblies

  4. SuggestedThe human assemblies contain about 133.6 Mbp of non-redundant segmental duplication (SD) per person, African-specific SD hotspots are enriched in pericentromeric regions, and gene copy gains such as extra intact AMY1 copies are resolved.

    BISER (v1.2.3): mean 133.6 Mbp SD (~4.7% of genome); total SD size vs genome size R2 = 0.9641, p < 2.2e-16. AMY1 gains in ten individuals (8 Asian, 2 African), e.g. 4 and 3 extra copies in people from Vietnam and Thailand; dense SD clusters at PRAME, OR, GAGE and MAGEA. Hotspot enrichment is described without a test statistic; no orthogonal copy-number check is reported. · Assembly of 35 human genomes by NextDenovo and…; Gene, gene duplications and repeat annotations; Segmental duplication (SD) analysis

  5. ClaimedThe discussion reports earlier NextDenovo assemblies of very large genomes, up to ~40 Gb lungfish and ~48 Gb krill, and near-chromosome contigs from ONT ultra-long reads.

    Cited applications, not benchmarked in this paper: Cycas panzhihuaensis ~10.5 Gb (contig N50 12 Mb), allohexaploid oat ~10.76 Gb (N50 75.27 Mb), African lungfish ~40 Gb (N50 1.60 Mb), Antarctic krill ~48 Gb (N50 178.99 kb), papaver ~4.59 Gb (N50 65.57 Mb), watermelon 3.69 Gb (11 longest contigs = 11 chromosomes). · Discussion

Methods, data and reproducibility

Correction-then-assembly pipeline for noisy ONT reads: k-mer score chain correction with iterative partial-order-alignment re-correction of low-score regions, two-round minimap2 overlapping and string-graph cleaning tuned for repeat nodes. Benchmarked against Canu, Necat, Flye and Wtdbg2 on human chr1, four plant/fly genomes and 35 newly sequenced human genomes.

  • Simulated ONT reads (~62X, N50 20.77 kb) from GRCh38 chromosome 1, generated with NanoSim · human
  • Real ONT ultra-long reads (~72X, N50 56.77 kb) from CHM13 chromosome 1 · human
  • A. thaliana ONT and short reads plus reference genome (BIG Data Center) · Arabidopsis thaliana · PRJCA005809; CRR302667 (ONT); CRR302670 (short reads); GWHBDNP00000001.1 (reference)
  • D. melanogaster ONT and short reads; NCBI reference genome · Drosophila melanogaster · SRR6702603, SRR6821890 (ONT); SRR6702604 (short reads); GCA_000001215.4 (reference)
  • O. sativa ONT, HiFi and short reads; hifiasm HiFi assembly of the same individual used as reference · Oryza sativa · SRR10948639-SRR10948642 (ONT); SRR10948643 (HiFi); SRR10948638 (short reads)
  • Z. mays ONT and short reads; NCBI reference genome · Zea mays · SRR12482959-SRR12482969 (ONT); SRR11870962 (short reads); GCA_014529475.1 (reference)
  • Genomes from blood of people of diverse ancestry (13 Africa, 6 East Asia, 4 Southeast Asia, 6 South Asia, 2 Middle East, 2 Europe, 1 Oceania, 1 America); ONT and MGISEQ2000 reads · human · n = 35 individuals · PRJCA006287 (GSA, NGDC)
  • 1000 Genomes Project SNP genotypes merged with the 35 genomes for principal component analysis · human
Methods & tools
NextDenovo, NextPolish, Canu, Necat, Flye, Wtdbg2, minimap2, NanoSim, QUAST, BUSCO, BISER, Liftoff
Reagents & environment
Comparator and analysis versions given (Canu v2.0, Necat v0.0.1, Flye v2.8, Wtdbg2 v2.5, QUAST v5.2.0, NanoSim v2.6.0, hifiasm v0.16.1, BISER v1.2.3, Liftoff v1.6.3, GATK v4.0.4.0); NextDenovo version not stated; commands and parameters in the Supplementary Information.
Assay conditions & compute
Subtasks can be spread across compute nodes, typically 32~64 GB peak memory each; most genomes finish in about a day on dozens of nodes.
Reproducibility gaps
Code, benchmarking-data and SD-analysis links were lost in extraction (the text says they are public but the URLs are missing); the CHM13 data source is also unlinked. Raw human data from Chinese individuals are under restricted access via a GSA-Human request.
Limitations (5)
  • Cannot yet produce haplotype-resolved assemblies from ONT data without trio binning, because of residual sequencing errors. · stated by the authors
  • Slower than assembly-then-correction tools such as Flye on large repeat-rich genomes because of the extra read-correction step. · stated by the authors
  • Reported run times may change with different parameter settings. · stated by the authors
  • Correction benchmark covers only chromosome 1 (one simulated and one real dataset), and the 35-genome human comparison is against Flye alone. · noted in this summary
  • SD hotspot and AMY1 copy-number findings rest on 35 individuals, with no statistical test for hotspot enrichment and no independent copy-number validation reported. · noted in this summary

Read this if you assemble large repeat-rich genomes or many human genomes from noisy ONT reads and want a correction-first assembler benchmarked against Canu, Necat, Flye and Wtdbg2.

Citations
110

Abstract

Long read sequencing data, particularly those derived from the Oxford Nanopore (ONT) sequencing platform, tend to exhibit a high error rate. Here, we present NextDenovo, a highly efficient error correction and assembly tool for noisy long reads, which achieves a high level of accuracy in genome assembly. NextDenovo can rapidly correct reads; these corrected reads contain fewer errors than other comparable tools and are characterized by fewer chimeric alignments. We applied NextDenovo to the assembly of high quality reference genomes of 35 diverse humans from across the world using ONT Nanopore long read sequencing data. Based on these de novo genome assemblies, we were able to identify the landscape of segmental duplications and gene copy number variation in the modern human population. The use of the NextDenovo program should pave the way for population-scale long-read assembly, thereby facilitating the construction of human pan-genomes, using Nanopore long read sequencing data.

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
THISChen '20Morisse '21Goodwin '15Coombe '21Holley '21Morisse '18
73% semantic matchBioinformatics
Ying Chen, Fan Nie, Shuang Xie+12 more·bioRxiv·Feb 1, 2020· 37 citations