De Novo Genome Assembly with Long Reads: A Practical Workflow

De novo genome assembly with long reads is the process of reconstructing a genome sequence from scratch using sequencing data from platforms such as Oxford Nanopore Technologies or Pacific Biosciences, without relying on a reference genome. This workflow guide covers the essential steps from raw read quality control through assembly, polishing, and evaluation, with concrete decisions for researchers, analysts, and life-science professionals managing genome projects. The practical outcome is a checklist you can apply to your own assembly pipeline, informed by published benchmarks and clinical genomics applications.

Scope and Context for Long-Read Assembly

Long-read sequencing technologies produce reads that are substantially longer than traditional short-read platforms, often spanning tens of kilobases. This length advantage directly addresses limitations of short-read assembly, particularly in complex repeat regions, structural variant detection, and haplotype phasing. The commercial release of third-generation sequencing technologies has stimulated development of new tools for assembling highly contiguous genome sequences with unprecedented accuracy across complex repeat regions [9]. For researchers deciding whether to pursue long-read assembly, the primary considerations are genome size, complexity, available computational resources, and the biological questions that the assembly must answer.

The practical context for long-read assembly extends beyond basic research. In clinical genomics, long-read sequencing has demonstrated potential to enhance diagnostic yield for rare monogenic diseases [5]. Studies of undiagnosed rare diseases show that long-read sequencing addresses many limitations of short-read approaches, detecting and clarifying additional disease-associated variants that may be missed by standard diagnostic workflows [6]. For plant genomics, workflows have been developed that produce high-quality chromosome-level assemblies in approximately seven days using modest laboratory and bioinformatic resources [10]. These examples establish that long-read assembly is not confined to large genome centers but is accessible to individual laboratories with appropriate planning.

Core Principles of Long-Read Assembly

Read Length and Coverage Requirements

The fundamental advantage of long reads is their ability to span repetitive elements and structural variations that confound short-read assemblers. Long-read sequencing data from nanopore platforms can achieve average coverage of approximately 36-fold and read N50 values of 32 kilobases from a single flow cell [5]. Coverage depth directly influences assembly contiguity and accuracy. For bacterial genomes, 100-fold coverage has been shown sufficient to maintain assembly accuracy under optimized conditions [19]. Larger and more complex genomes generally require higher coverage to achieve comparable completeness.

Error Profiles of Long-Read Platforms

Long reads typically have higher error rates relative to short reads [7]. If left unaddressed, subsequent genome assemblies may exhibit high base error rates that compromise the reliability of downstream analysis [7]. The error profiles differ between platforms and even between flow cell versions. For Oxford Nanopore data, R10.4.1 flow cells with super accurate basecalling produce substantially fewer errors than R9.4.1 flow cells, particularly in repetitive sequences [19]. Understanding your platform's error characteristics informs the choice of assembly and polishing tools.

Assembly Graph Concepts

Most long-read assemblers operate on assembly graphs, structures that represent the relationships between sequencing reads. The assembler identifies overlaps between reads and constructs a graph where contigs represent paths through the graph. Progressive error correction with consensus refinement is a strategy employed by several assemblers to generate near-complete assemblies [13]. The choice of assembler determines how errors are handled during graph construction and how the final consensus sequence is derived.

Practical Workflow for De Novo Assembly

Step 1: Read Quality Control and Preprocessing

Before assembly begins, raw sequencing reads require quality assessment and filtering. The specific parameters depend on your sequencing platform and basecalling approach. For nanopore data, filtering reads by length and quality score is a standard preprocessing step. One optimized workflow used reads of at least 1 kilobase with quality scores of 10 or higher before assembly [19]. Read trimming removes adapter sequences and low-quality bases that could introduce errors into the assembly graph.

Quality control decisions should be recorded systematically. Document the total number of reads, total bases, read N50, mean quality score, and the number of reads removed during filtering. These metrics provide the baseline for evaluating assembly performance and troubleshooting failures.

Step 2: Selecting an Assembly Tool

The choice of assembler substantially influences assembly quality [13]. Benchmarks of eleven long-read assemblers on bacterial genomes found that assemblers employing progressive error correction with consensus refinement, notably NextDenovo and NECAT, consistently generated near-complete single-contig assemblies with low misassemblies [13]. Flye offered a strong balance of accuracy and contiguity, although it was sensitive to corrected input [13]. Canu achieved high accuracy but produced fragmented assemblies and required the longest runtimes [13]. Ultrafast tools such as Miniasm and Shasta provided rapid draft assemblies but were highly dependent on preprocessing and required polishing to achieve completeness [13].

For bacterial genomes, a fully graphical user interface-based workflow using the Galaxy platform has been validated for de novo assembly using Nanopore data alone without short-read correction [19]. This workflow used Flye for assembly followed by Medaka polishing with the bacterial methylation model, achieving an error rate of 0.0007% with R10.4.1 data [19]. This demonstrates that command-line expertise is not an absolute requirement for producing high-quality assemblies.

Step 3: Assembly Polishing

Polishing is the process of correcting base-level errors in genome assemblies [15]. Long reads typically have higher error rates relative to short reads, and if left unaddressed, assemblies may exhibit high base error rates that compromise downstream analysis [7]. Several specialized error correction tools have emerged, employing a range of algorithms and strategies to improve base quality [7].

Polishing approaches face notable trade-offs. Alignment-based methods achieve high accuracy but incur long run times, while alignment-free k-mer-based tools are scalable but struggle in regions with dense errors [15]. Machine learning-based polishers often perform well on specific platforms but require read-to-assembly alignments [15]. The choice of polishing tool should consider your computational resources, the error profile of your sequencing platform, and the accuracy requirements of your downstream analyses.

Targeted polishing offers a resource-efficient alternative to whole-genome polishing. GoldPolish-Target is a modular targeted sequence polishing pipeline that isolates and polishes user-specified assembly loci [7]. Experiments using Drosophila melanogaster and Homo sapiens datasets demonstrated that GoldPolish-Target can reduce insertion/deletion and mismatch errors by up to 49.2% and 55.4% respectively, achieving base accuracy values upwards of 99.9% [7]. This polishing accuracy is comparable to current state-of-the-art tools while exhibiting up to 27-fold shorter run times and consuming 95% less memory [7].

Step 4: Assembly Evaluation

Assembly evaluation requires multiple complementary metrics. Contiguity metrics include N50, total length, and contig count [13]. Completeness is assessed using Benchmarking Universal Single-Copy Orthologs (BUSCO), which checks for the presence of conserved single-copy genes expected in the genome [13]. Base-level accuracy can be measured using quality scores derived from read-to-assembly alignments or from k-mer-based methods.

For clinical applications, assembly accuracy has direct implications for variant detection. Long-read assemblies have been used to detect additional rare functionally annotated variants, including structural variants and tandem repeats, compared to short-read sequencing [5]. In one study, long-read sequencing completely phased 87% of protein-coding genes [5]. The quality of your assembly determines the reliability of these downstream analyses.

At a Glance: Assembly Workflow Decision Table

Options and Tradeoffs in Assembly Tools

Assembler Comparison for Bacterial Genomes

The benchmark of eleven long-read assemblers provides concrete guidance for tool selection [13]. NextDenovo and NECAT consistently generated near-complete single-contig assemblies with low misassemblies and stable performance across preprocessing types [13]. Flye offered a strong balance of accuracy and contiguity, although it was sensitive to corrected input [13]. Canu achieved high accuracy but produced fragmented assemblies of 3 to 5 contigs and required the longest runtimes [13]. Unicycler reliably produced circular assemblies but with slightly shorter contigs than Flye or NextDenovo [13]. Ultrafast tools such as Miniasm and Shasta provided rapid draft assemblies yet were highly dependent on preprocessing and required polishing to achieve completeness [13]. HINGE and wtdbg2 underperformed in this benchmark [13].

Platform-Specific Considerations

The choice between Oxford Nanopore and Pacific Biosciences platforms affects both error profiles and assembly strategies. Advanced long-read sequencing technologies from both providers are finding wide use in de novo genome sequencing projects [7]. For nanopore data, the flow cell version substantially impacts assembly accuracy. R10.4.1 flow cells with Dorado super accurate basecalling achieved an error rate of 0.0007% in a bacterial assembly workflow, while R9.4.1 yielded an error rate of 0.1989% [19]. R9.4.1 data contained more indels, particularly in repetitive sequences, whereas R10.4.1 data exhibited a higher frequency of residual SNP differences, with most remaining discrepancies enriched in homopolymeric tracts or other low-complexity regions [19].

Hybrid Assembly Approaches

Hybrid assembly integrates long reads with short-read data to combine the advantages of both technologies. Short-read RNA sequencing remains a cornerstone for transcriptome profiling but is limited in reconstructing full-length transcripts [14]. Long-read RNA sequencing spans entire transcripts and resolves complex structures but is hindered by high error rates [14]. Hybrid de novo RNA assembly pipelines integrate the accuracy of short reads with the structural resolution of long reads to produce more complete transcriptome assemblies [14]. For genome assembly, hybrid approaches can use short reads for polishing long-read assemblies, although optimized workflows can achieve high accuracy with nanopore data alone [19].

Records and Measurements for Assembly Projects

Essential Documentation

Maintain a laboratory notebook or electronic record that captures the following for each assembly project:

• Sequencing platform, flow cell version, and basecalling model
• Read filtering parameters and the number of reads removed at each step
• Assembler name, version, and all non-default parameters
• Polishing tool, version, and target regions if targeted polishing was used
• Assembly metrics including N50, total length, contig count, and BUSCO completeness
• Computational resources used including runtime and peak memory

This documentation supports reproducibility and troubleshooting. When assembly problems arise, the record allows you to identify which parameter changes affected the outcome.

Quality Metrics and Their Interpretation

Assembly quality is multidimensional. Contiguity metrics such as N50 describe the length distribution of contigs but do not measure accuracy. Completeness metrics such as BUSCO assess whether expected genes are present but do not detect misassemblies that preserve gene content. Base accuracy metrics such as quality scores measure per-base error rates but may not capture structural errors. A comprehensive evaluation requires all three categories.

For clinical applications, the accuracy requirements are stringent. Long-read sequencing has been used to detect additional de novo variants and to distinguish postzygotic mosaic variants from prezygotic de novos [5]. These applications demand high base accuracy and correct haplotype phasing. The quality score of an assembly, often reported as QV, provides a summary measure of base accuracy. Assemblies with QV values above 50 are considered high quality for many applications [11].

Common Failure Patterns in Long-Read Assembly

Insufficient Coverage

Low coverage leads to fragmented assemblies with gaps in repetitive regions. The required coverage depends on genome size and complexity. For bacterial genomes, 100-fold coverage appeared sufficient to maintain accuracy under optimized conditions [19]. Larger genomes with high repeat content require proportionally more data. If your assembly shows excessive fragmentation, check whether coverage was adequate for the genome size.

Inadequate Read Filtering

Including short or low-quality reads in the assembly can introduce errors and complicate graph construction. One optimized workflow filtered reads to at least 1 kilobase with quality scores of 10 or higher [19]. If your assembly contains many small contigs or shows elevated error rates, revisit your filtering parameters.

Incorrect Polishing Strategy

Skipping polishing or using an inappropriate polishing tool can leave high error rates in the final assembly. Long reads typically have higher error rates relative to short reads, and if left unaddressed, assemblies may exhibit high base error rates that compromise downstream analysis [7]. If your assembly has low base accuracy, consider whether polishing was performed correctly and whether the polishing tool matches your sequencing platform.

Parameter Sensitivity

Some assemblers are sensitive to input preprocessing. Flye was sensitive to corrected input in benchmarking studies [13]. If your assembly quality changes dramatically with different preprocessing approaches, document the sensitivity and standardize your workflow.

Limitations and Interpretation Boundaries

Error Rates in Complex Regions

Even polished assemblies retain elevated error rates in certain regions. Homopolymeric tracts and other low-complexity regions remain challenging for nanopore sequencing [19]. These regions may require targeted validation with orthogonal methods. When interpreting variants in such regions, exercise caution and consider confirmatory assays.

Assembly Versus Reference-Based Analysis

De novo assembly provides advantages for detecting structural variants and resolving complex regions, but it is computationally intensive. Reference-based analysis may be more appropriate for projects where the reference genome is closely related to the sample. Long-read sequencing can be used for both approaches, and the choice depends on your research question and resources.

Computational Resource Requirements

Assembly and polishing are computationally demanding. Some polishing tools require multi-day run times and substantial memory [15]. Machine learning-based polishers can reduce resource requirements while maintaining accuracy [15]. Assess your computational infrastructure before selecting tools, and consider cloud computing options for large genomes.

Safety and Regulatory Context

Data Sharing and Privacy

Genomic data from human subjects is subject to privacy regulations and institutional policies. The NIH Genomic Data Sharing Policy establishes expectations for data sharing, including protections for participant privacy [3]. If your assembly project involves human data, ensure compliance with applicable policies before depositing or sharing sequence data.

Data Repositories and Standards

Public repositories such as NCBI provide infrastructure for depositing and accessing genomic data [2]. The FAIR Guiding Principles describe best practices for making data findable, accessible, interoperable, and reusable [4]. Following these principles when publishing assemblies and associated metadata supports scientific reproducibility and secondary analysis.

Training and Competency

Bioinformatics training resources are available through organizations such as EMBL-EBI [1]. Researchers performing genome assembly should have appropriate training in the tools and methods they use. For clinical applications, additional validation and quality assurance procedures may be required.

Professional Escalation Criteria

When to Seek Expert Assistance

Consider consulting a bioinformatics specialist or core facility when:

• Your assembly produces unexpected results that persist after parameter optimization
• You are working with a genome of unusual size or complexity
• Your project has clinical or regulatory implications that require validated workflows
• You lack the computational resources to complete assembly and polishing in a reasonable timeframe

When to Reconsider the Workflow

Reassess your approach if:

• Assembly quality metrics fall below the thresholds required for your downstream analyses
• Polishing does not achieve the expected accuracy improvements
• Runtime or memory requirements exceed your available resources by a substantial margin
• New versions of assembly or polishing tools offer significant improvements over your current workflow

Frequently Asked Questions

What coverage depth is needed for long-read de novo assembly?

Coverage requirements depend on genome size and complexity. For bacterial genomes, 100-fold coverage appeared sufficient to maintain assembly accuracy under optimized conditions [19]. Larger and more complex genomes generally require higher coverage. A nanopore sequencing study of a rare disease cohort achieved approximately 36-fold average coverage per sample with a 32-kilobase read N50 from a single flow cell [5]. For repetitive or polyploid genomes, consult published benchmarks for your organism type.

Which assembler should I choose for bacterial genomes?

Benchmarking studies provide concrete guidance. NextDenovo and NECAT consistently generated near-complete single-contig assemblies with low misassemblies [13]. Flye offered a strong balance of accuracy and contiguity [13]. Canu achieved high accuracy but produced fragmented assemblies and required the longest runtimes [13]. Ultrafast tools such as Miniasm and Shasta provided rapid draft assemblies but required polishing to achieve completeness [13]. Your choice should balance accuracy, contiguity, and available computational resources.

Is polishing always necessary after long-read assembly?

Polishing is critical for ensuring accuracy in downstream analyses such as variant calling, gene annotation, and clinical genomics applications [15]. Long reads typically have higher error rates relative to short reads, and if left unaddressed, assemblies may exhibit high base error rates [7]. Even with recent advances in long-read sequencing technologies, maintaining high base-level accuracy remains challenging due to the still appreciable errors associated with certain long-read sequencing technologies [15]. Skipping polishing is only advisable for draft assemblies intended for preliminary analysis.

Can I assemble a genome using nanopore data alone without short-read correction?

Yes, optimized workflows can achieve high accuracy with nanopore data alone. A fully graphical user interface-based workflow using the Galaxy platform assembled Haemophilus influenzae genomes using Nanopore data alone without short-read correction [19]. With R10.4.1 flow cells, Dorado super accurate basecalling, read filtering, Flye assembly, and Medaka polishing, the error rate was 0.0007% [19]. This demonstrates that short-read correction is not mandatory when using current nanopore chemistry and appropriate polishing.

What is the difference between whole-genome and targeted polishing?

Whole-genome polishing corrects errors across the entire assembly, which can be computationally expensive. Targeted polishing isolates and polishes user-specified assembly loci, offering a resource-efficient means for polishing targeted regions of draft genomes [7]. GoldPolish-Target can reduce indel and mismatch errors by up to 49.2% and 55.4% respectively while exhibiting up to 27-fold shorter run times and consuming 95% less memory compared to whole-genome approaches [7]. Targeted polishing is appropriate when only specific regions require high accuracy.

How do I evaluate the quality of my assembly?

Assembly evaluation requires multiple complementary metrics. Contiguity metrics include N50, total length, and contig count [13]. Completeness is assessed using Benchmarking Universal Single-Copy Orthologs (BUSCO) [13]. Base-level accuracy can be measured using quality scores derived from read-to-assembly alignments or k-mer-based methods. For clinical applications, additional validation may be required to confirm variant calls.

What are the main advantages of long-read assembly for clinical genomics?

Long-read sequencing can capture variants inaccessible to short-read sequencing, facilitate long-range mapping and phasing, and provide haplotype-resolved methylation profiling [5]. Studies of undiagnosed rare diseases show that long-read sequencing addresses many limitations of short-read approaches, detecting and clarifying additional disease-associated variants [6]. Key areas of added diagnostic yield include detection and resolution of structural variants, detection and characterization of tandem repeat expansions, coverage of regions of high sequence similarity, and variant phasing [6].

How should I document my assembly workflow for reproducibility?

Maintain records of sequencing platform, flow cell version, basecalling model, read filtering parameters, assembler name and version, polishing tool and version, assembly metrics, and computational resources used. Following the FAIR Guiding Principles makes your data findable, accessible, interoperable, and reusable [4]. Public repositories such as NCBI provide infrastructure for depositing and accessing genomic data [2].

Related Bioinformatics Guides

• Long-Read Genome Assembly and Polishing Strategies
• Long Read Metagenomic Assembly: Structural Analysis and Computational Methodologies in Bioinformatics
• Transcriptome Assembly Without a Reference Genome
• Long-Read Sequencing Technologies: PacBio and Oxford Nanopore
• FASTQ File Format: Decoding Phred Quality Scores and Quality Control Workflows

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Advancing long-read nanopore genome assembly and accurate variant calling for rare disease detection.. American journal of human genetics, 2025.
• The additional diagnostic yield of long-read sequencing in undiagnosed rare diseases.. Genome research, 2025.
• GoldPolish-target: targeted long-read genome assembly polishing.. BMC bioinformatics, 2025.
• Advancing long-read nanopore genome assembly and accurate variant calling for rare disease detection.. medRxiv : the preprint server for health sciences, 2024.
• Tools and Strategies for Long-Read Sequencing and De Novo Assembly of Plant Genomes.. Trends in plant science, 2019.
• LeafGo: Leaf to Genome, a quick workflow to produce high-quality de novo plant genomes using long-read sequencing technology.. Genome biology, 2021.
• Pangenome discovery of missing autism variants.. medRxiv : the preprint server for health sciences, 2025.
• GALA: a computational framework for de novo chromosome-by-chromosome assembly with long reads.. Nature communications, 2023.
• Benchmarking long-read assembly tools and preprocessing strategies for bacterial genomes: A case study on &lt,i&gt,E. coli&lt,/i&gt, DH5α.. 2025.
• HyDRA: A pipeline for integrating long- and short-read RNAseq data for custom transcriptome assembly.. 2026.
• AIEdit: Alignment-free genome assembly polisher trained on spaced seed match patterns.. 2026.
• &lt,i&gt,StrainCascade&lt,/i&gt,: An automated, modular workflow for high-throughput long-read bacterial genome reconstruction and characterization.. 2026.
• Nonhybrid, finished microbial genome assemblies from long-read SMRT sequencing data. Nature Methods, 2013.
• Methods for De-novo Genome Assembly. 2020.
• Optimized De novo assembly of Haemophilus influenzae using oxford nanopore sequencing without short-read data on a graphical user interface-based galaxy platform: Accurate detection of MLST, PBP3 mutations, and phylogenetic clustering.. Diagnostic microbiology and infectious disease, 2026.
• Ultra-low biomass sequencing workflow (LBV-Seq) enables de novo metagenomic reconstruction of DNA and RNA viral genomes. bioRxiv, 2026.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.