Nanopore sequencing and assembly of a human genome with ultra-long reads
Miten JainiD, Sergey KoreniD, Karen H. MigaiD, Josh Quick, Arthur C Rand, Thomas A. SasaniiD, John R. TysoniD, Andrew David BeggsiD, Alexander DiltheyiD, Ian T. FiddesiD, Sunir Malla, Hannah Marriott, Tom NietoiD, Justin O’GradyiD, Hugh E. OlseniD, Brent S. PederseniD, Arang RhieiD, Hollian Richardson, Aaron R. QuinlaniD, Terrance P. SnutchiD, Louise J. TeeiD, Benedict PateniD, Adam M. PhillippyiD, Jared T. SimpsoniD, Nicholas James LomaniD, Matthew LooseiD
Abstract
Nanopore sequencing is a promising technique for genome sequencing due to its portability, ability to sequence long reads from single molecules, and to simultaneously assay DNA methylation. However until recently nanopore sequencing has been mainly applied to small genomes, due to the limited output attainable. We present nanopore sequencing and assembly of the GM12878 Utah/Ceph human reference genome generated using the Oxford Nanopore MinION and R9.4 version chemistry. We generated 91.2 Gb of sequence data ([~]30x theoretical coverage) from 39 flowcells. De novo assembly yielded a highly complete and contiguous assembly (NG50 [~]3Mb). We observed considerable variability in homopolymeric tract resolution between different basecallers. The data permitted sensitive detection of both large structural variants and epigenetic modifications. Further we developed a new approach exploiting the long-read capability of this system and found that adding an additional 5x-coverage of ultra-long reads (read N50 of 99.7kb) more than doubled the assembly contiguity. Modelling the repeat structure of the human genome predicts extraordinarily contiguous assemblies may be possible using nanopore reads alone. Portable de novo sequencing of human genomes may be important for rapid point-of-care diagnosis of rare genetic diseases and cancer, and monitoring of cancer progression. The complete dataset including raw signal is available as an Amazon Web Services Open Dataset at: https://github.com/nanopore-wgs-consortium/NA12878.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 1,892
- FWCI
- 90.95
- Field percentile
- top 1%
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · xhigh effort · 7 Nov 2026 · from the full text of version 1 (20 Apr 2017) · CC BY 4.0
Nanopore-only sequencing of GM12878 across 39 MinION flowcells (91.2 Gb) yielded an assembly with NG50 ∼3 Mb; adding 5× ultra-long reads raised NG50 to 6.4 Mbp.
Short reads (∼25–300 bp for Illumina) cannot bridge the repeats and segmental duplications that make up nearly half the human genome, and limited MinION output had confined nanopore assembly to microbes and small eukaryotes. This work shows de novo assembly of a human genome from nanopore reads alone, with native methylation and structural variant calling, and an ultra-long read protocol that more than doubles contiguity.
Key results
ShownA 30× nanopore-only Canu assembly of GM12878 was contiguous but error-prone before polishing; Illumina-based Pilon polishing lifted accuracy, and comparative annotation recovered most GENCODE genes with very few split across contigs.
2,886 contigs, NG50 3 Mbp; identity 95.20% vs GRCh38 and 95.74% vs independent Illumina data; 99.88% after Pilon; 58,338 genes annotated (96.4% of GENCODE V24), only 857 (0.1%) on two or more contigs; 899 structural differences vs 692 for a prior PacBio GM12878 assembly. · Assembly of the dataset
ShownSaturating a modified Oxford Nanopore Rapid Kit with high molecular weight DNA gave ultra-long reads; adding this low coverage more than doubled assembly contiguity and placed the MHC in one contig.
Extra 5× coverage with read N50 99.7 kb; longest full-length mapped read 882 kb spanning 993 kb of reference; combined 35× assembly NG50 6.4 Mbp vs 3 Mbp for the 30× assembly; MHC contained in a single 15 Mbp contig. · Ultra-long reads to improve assembly contiguity; Discussion; Figure legends
ShownOn chromosome 20, basecaller choice drove homopolymer resolution and consensus accuracy, and signal-level polishing produced the highest accuracy reached from nanopore data alone.
Median read identity 82.43% (Metrichor), 85.50% (Nanonet), 86.05% (Scrappie); Metrichor and Nanonet capped homopolymers at 5 bp; chr20 assembly accuracy rose from 95.74% to 97.80% with Scrappie reads, 99.44% after nanopolish, and 99.96% (QV34) with Illumina added. · Sequence analysis and base caller evaluation; Assembly of the dataset; Discussion; Base call analysis
ShownNanopore alignments recovered most known structural variants and gave high overall SNP genotype accuracy on chromosome 20, though variant-site calls were less accurate.
With 39 flowcells, ∼93% of 2,435 high-confidence SVs recovered at ∼6% false-positive rate (6.4% FDR over 1000 permutations); agreement with Illumina 82% het, 91% hom-alt. SNPs: 99.16% correct (778,412 of 784,998), 91.40% at called variant sites (50,814 of 55,595), 94.83% at platinum variants (50,814 of 53,582). · Measuring SNP and SV genotyping sensitivity; Estimating SV genotyping sensitivity
ShownCpG 5-methylcytosine called directly from native nanopore signal on chromosome 20 tracked bisulfite sequencing, with nanopolish agreeing more closely than SignalAlign.
Correlation with ENCODE bisulfite data (ENCFF835NTC): r = 0.895 for nanopolish and 0.779 for SignalAlign; both models trained on Zymo unmethylated and fully CpG-methylated WGA control DNA. · Native 5-methyl cytosine detection; Generating a controlled set of methylated…
Methods, data and reproducibility
PCR-free GM12878 DNA from five laboratories was sequenced on MinION R9/R9.4 flowcells with 1D ligation and modified rapid kits, assembled with Canu, and polished with Pilon (Illumina) or nanopolish (signal). Reads were also tested for basecaller bias, SV and SNP genotyping, CpG methylation, and repeat-based contiguity modelling.
- GM12878 native DNA sequenced with standard 1D ligation and rapid libraries on MinION R9/R9.4 (91.2 Gb, read N50 10,589 bp) · human (GM12878 cell line) · n = 39 flowcells; 14,183,584 reads
- Ultra-long reads from phenol/chloroform-extracted GM12878 DNA with a modified SQK-RAD002 rapid kit (read N50 99.7 kb), basecalled with Albacore · human (GM12878 cell line) · n = 5× coverage
- Whole-genome Illumina data for GM12878 used for Pilon polishing · human · SRA:ERP001229
- High-confidence GM12878 deletions/duplications genotyped in Platinum Genomes Illumina paired-end data · human · n = 2,435 SVs · ENA run ERR194147
- Genome In a Bottle NA12878 Illumina HiSeq downsampled subset for reference-free identity estimates · human · n = 30-fold subset
- 1000 Genomes phase 3 non-singleton chr20 SNPs evaluated against Illumina Platinum genotype calls · human · n = 784,998 sites
- ENCODE GM12878 bisulfite sequencing used as methylation comparison set on chr20 · human · ENCFF835NTC; ENCSR890UQO
- Zymo methylation control standards (cat. D5013): unmethylated and fully CpG-methylated WGA DNA for model training · n = two flowcells
- Methods & tools
- Oxford Nanopore MinION R9/R9.4 flowcells (SQK-LSK108 ligation, SQK-RAD002 rapid kits), Basecallers: Metrichor, Nanonet v2.0, Scrappie v0.2.7, Albacore 0.8.4, Canu v1.4, Pilon, nanopolish, SignalAlign and marginAlign (ported to Toil), BWA-MEM 0.7.12-r1039, SVTyper (modified for nanopore bleed-through alignments), MUMmer v3.23 and Assemblytics, Comparative Annotation Toolkit (CAT) with AUGUSTUS, FreeBayes v1.0.2, WhatsHap
- Code
- http://github.com/tomsasani/svtyperhttps://github.com/ArtRand/toil-signalAlignhttps://github.com/ArtRand/toil-marginAlignhttps://github.com/nickloman/poredbhttps://github.com/minoTour/minoTourhttps://github.com/ComparativeGenomicsToolkit/Comparative-Annotation-Toolkit
- Data & accessions
- https://github.com/nanopore-wgs-consortium/NA12878http://hgdownload.soe.ucsc.edu/goldenPath/hg38/database/
- Software & environment
- Canu v1.4 r8006 (20×/30×) and r8120 (35×); BWA-MEM 0.7.12-r1039; MUMmer v3.23; FreeBayes v1.0.2; Nanonet v2.0; Scrappie v0.2.7; Albacore 0.8.4; GATK 3.7; CAT commit c9503e7; analysis on CLIMB servers and Toil cloud pipelines.
- Assay conditions & compute
- Canu took 40K CPU hours (20×), 62K CPU hours (30×; about fourfold slower than comparable PacBio) and 151K CPU hours (35× with ultra-long reads); unfiltered Scrappie reads stalled BWA-MEM for >30 days on 16-cores.
- Reproducibility gaps
- Albacore 0.8.4 and the modified MinKNOW run scripts are available only through the Oxford Nanopore user community; Scrappie is cited as ONT software without a public link; the GIAB Illumina subset is given only as an ftp:// path.
Limitations (6)
- Unpolished nanopore-only consensus is error-prone, dominated by deletions from truncated homopolymers and low-complexity sequence; Illumina polishing cannot correct the most repetitive regions because short reads map poorly there, and a deletion bias remains after polishing. · stated by the authors
- Nanopore genotyping trails short-read platforms, chiefly in separating heterozygous from homozygous alleles; the main SNP error was calling homozygous-reference sites heterozygous (3,217 errors). · stated by the authors
- The data-handling workflow (one file per read, custom tracking tools, cloud pipelines) is not convenient for other groups to repeat; BWA-MEM could not finish aligning the ultra-long reads, and the longest reads exceed BAM CIGAR limits. · stated by the authors
- Canu's trimming step was not designed for extremely long, error-prone reads, which pushed the 35× assembly to 151K CPU hours; the prediction that 30× of ultra-long reads would exceed 40Mb NG50 was not tested. · stated by the authors
- The basecaller comparison, signal-level polishing (99.44%, 99.96%), SNP genotyping and methylation calling were assessed on chromosome 20 only; the whole-genome assembly used Metrichor reads, so these accuracies do not apply to it. · noted in this summary
- Some figures disagree within the text: the chr20 baseline before Scrappie is 95.74% in Results but 95.50% in Discussion, and the projected NG50 for 30× ultra-long reads is over 40Mb in Discussion but over 30 Mbp in the Figure 5 legend. · noted in this summary
Who should read:Read this if you are planning long-read de novo assembly of human genomes or want GM12878 benchmarks of nanopore basecallers, signal-level polishing, SV genotyping and CpG methylation calling.
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
- Efficient de novo assembly of eleven human genomes using PromethION sequencing and a novel nanopore toolkit
Kishwar Shafin, Trevor Pesout, … Benedict Paten·Nature BiotechnologyPeer-reviewed·May 4, 2020
- Gapless assembly of complete human and plant chromosomes using only nanopore sequencing
Sergey Koren, Zhigui Bao, … Adam M. Phillippy·Genome ResearchPeer-reviewed·Nov 1, 2024
- Scaffolding and Completing Genome Assemblies in Real-time with Nanopore Sequencing
Minh Duc Cao, Son Hoang Nguyen, … Lachlan J M Coin·bioRxiv·May 22, 2016
- A single-molecule nanopore sequencing platform
Jiayuan Zhang, Yuning Zhang, … Yuliang Dong·bioRxiv·Aug 19, 2024
- A complete bacterial genome assembled de novo using only nanopore sequencing data
Nicholas James Loman, Joshua Quick, Jared T Simpson·Nature MethodsPeer-reviewed·Aug 1, 2015
- Oxford Nanopore Sequencing, Hybrid Error Correction, and de novo Assembly of a Eukaryotic Genome
Sara Goodwin, James Gurtowski, … W Richard McCombie·Genome ResearchPeer-reviewed·Oct 7, 2015