- Long-read genome assembly is the process of reconstructing genomic sequences using DNA sequencing reads that are substantially longer than conventional short reads. These reads may range from several thousand bases to tens or even hundreds of thousands of bases, depending on the sequencing technology and library preparation method. Long-read sequencing is particularly valuable for assembling complex genomes because longer reads can span repetitive regions, connect distant genomic segments, and resolve structural variations that are difficult to reconstruct using short-read data alone. It has become an important approach in microbial genomics, plant and animal genome research, comparative genomics, population genetics, and evolutionary biology.
- The primary objective of long-read genome assembly is to reconstruct long, continuous, and biologically accurate DNA sequences from sequencing reads. Unlike short-read assembly, which often relies on relatively short overlapping sequences or k-mer-based graph construction, long-read assembly can use extensive overlaps between reads to identify genomic relationships. Long reads may span entire repetitive elements, duplicated regions, structural variants, and other complex genomic features. This can substantially improve assembly contiguity and reduce the fragmentation commonly observed in short-read assemblies.
- Two major sequencing technologies are widely used to generate long reads: Pacific Biosciences (PacBio) and Oxford Nanopore Technologies (ONT). PacBio sequencing uses single-molecule real-time sequencing, while Oxford Nanopore sequencing determines DNA sequence by measuring changes in electrical current as nucleic acid molecules pass through nanopores. Both technologies generate long reads, but their sequencing mechanisms, error profiles, read characteristics, and computational requirements differ.
- PacBio technology includes Continuous Long Reads (CLR) and circular consensus sequencing (CCS). In traditional CLR sequencing, DNA molecules are sequenced as long reads, but individual reads may contain relatively high error rates. CCS sequencing takes advantage of circular DNA templates that allow the sequencing instrument to read the same DNA molecule multiple times. These repeated observations are combined to produce a consensus sequence. High-quality CCS reads are commonly referred to as HiFi reads. PacBio HiFi reads combine relatively long read lengths with high consensus accuracy and are widely used for de novo genome assembly, variant detection, haplotype resolution, and comparative genomics.
- Oxford Nanopore sequencing generates reads by passing DNA or RNA molecules through nanopores embedded in a membrane. As nucleotides move through the pore, they produce characteristic changes in electrical current. Basecalling algorithms interpret these signals and convert them into nucleotide sequences. Oxford Nanopore platforms can generate long and ultra-long reads, sometimes extending beyond hundreds of kilobases under suitable experimental conditions. These reads are particularly useful for spanning long repetitive regions, assembling complex genomes, identifying structural variations, and resolving chromosome-scale genomic structures. Nanopore sequencing can also provide information about certain DNA modifications through characteristic signal patterns, although modification detection requires appropriate models and validation.
- Long-read sequencing data are commonly stored in FASTQ format, which contains nucleotide sequences and their associated quality scores. Depending on the platform and workflow, raw signal data may also be retained. Oxford Nanopore raw signal data are commonly stored in POD5 files, while older workflows may use FAST5. PacBio workflows may produce subreads, circular consensus reads, and HiFi reads in formats such as BAM and FASTQ. Retaining raw signal and intermediate files can be important for reproducibility, re-basecalling, and quality assessment.
- The first stage of long-read genome assembly is sequencing data quality assessment. Researchers examine read length distributions, total sequencing yield, quality scores, nucleotide composition, adapter contamination, and possible sample contamination. Read length is particularly important because long reads can span repeated regions and connect genomic segments that would otherwise remain separated. However, longer reads are not automatically more useful if their sequence accuracy is poor or if the data contain substantial contamination. A balanced evaluation of read length, accuracy, coverage, and biological origin is therefore essential.
- Common long-read quality-control tools include NanoPlot, NanoStat, pycoQC, and MultiQC for Oxford Nanopore data. Tools such as SMRT Link and related PacBio utilities are used to process and assess PacBio sequencing data. The appropriate software depends on the platform, file type, and sequencing workflow. Quality-control reports should be examined before assembly to identify unusual read-length distributions, low-quality data, insufficient yield, and technical problems.
- Read length is commonly summarized using the mean, median, and N50 statistics. Read N50 is the length at which 50% of all sequenced bases are contained in reads of that length or longer. It provides information about the distribution of read lengths but does not describe sequence accuracy or genome completeness. A dataset with a high read N50 may still be unsuitable for assembly if its coverage is insufficient or its reads contain substantial errors.
- Sequencing coverage is another important factor in long-read assembly. It estimates how many times each genomic position is represented by sequencing reads on average. It can be calculated using the following expression:
[
C = \frac{N \times L}{G}
]
- where (C) is the approximate coverage, (N) is the number of reads, (L) is the average read length, and (G) is the estimated genome size. For example, if a sequencing project generates 2 million reads with an average length of 15,000 bases, the total sequencing output is approximately 30 billion bases. For a genome of 1 billion bases, this represents an estimated average coverage of 30-fold.
- Coverage requirements depend on sequencing accuracy, genome complexity, ploidy, heterozygosity, repeat content, and the assembler being used. PacBio HiFi assemblies may achieve high accuracy with moderate coverage under suitable conditions, while noisier long-read datasets may require greater coverage and additional consensus polishing. Extremely high coverage is not always necessary and may increase computational demands or complicate the handling of redundant and highly similar reads. The required depth should therefore be determined according to the sequencing technology and assembly objective.
- Long-read genome assembly commonly uses overlap-based approaches, including overlap-layout-consensus (OLC), string graph methods, and related graph-based algorithms. These methods identify overlaps between reads, construct relationships among overlapping sequences, and use the resulting graph to reconstruct longer genomic sequences. Unlike traditional short-read de Bruijn graph assembly, which represents fixed-length k-mers, overlap-based methods can work directly with long reads and their sequence relationships.
- The first major step is overlap detection. The assembler identifies regions of similarity between different reads, including overlaps involving reverse-complement orientations. Because a genome may contain millions of reads, comparing every read against every other read directly would be computationally expensive. Modern assemblers use efficient indexing strategies, minimizers, sketches, seed-and-extend methods, and alignment algorithms to identify likely overlaps before performing more detailed comparisons.
- The next step is graph construction and layout. Reads or read segments are represented as nodes or graph elements, while overlaps define possible connections. The assembler examines these relationships to determine how reads may be ordered and oriented along the genome. Repetitive regions can create multiple possible paths, while sequencing errors may generate false connections. Graph simplification, repeat handling, and coverage analysis help identify plausible genomic paths.
- After a suitable layout has been established, consensus generation reconstructs the nucleotide sequence supported by overlapping reads. Multiple reads covering the same genomic region provide evidence for the correct base at each position. Consensus algorithms use sequence alignments, quality information, error models, and statistical methods to distinguish genuine genomic variation from sequencing errors. The resulting sequences are assembled into contigs, which may subsequently be scaffolded or further corrected.
- Several long-read assemblers are widely used in bioinformatics. Canu is designed for long-read assembly and supports error correction, trimming, and assembly workflows. Flye is commonly used for long-read genome assembly and supports several long-read data types. Hifiasm is widely used for assembling PacBio HiFi reads and can generate phased or haplotype-resolved assemblies under suitable conditions. Shasta is designed for efficient assembly of long-read data, particularly Oxford Nanopore reads. Other tools include wtdbg2, HiCanu, and Raven. The choice of assembler depends on sequencing technology, read accuracy, genome size, ploidy, heterozygosity, and computational resources.
- Canu can be used for long-read assembly with suitable Oxford Nanopore or PacBio datasets. A basic example is:
canu \
-p genome \
-d canu_output \
genomeSize=500m \
-nanopore reads.fastq.gz
- In this example, -p specifies the output prefix, -d defines the output directory, and genomeSize provides the estimated genome size. The -nanopore option identifies the input as Oxford Nanopore data. For PacBio datasets, the corresponding input option should be selected according to the read type and Canu version. The estimated genome size should be adjusted to the organism being assembled.
- Flye is another commonly used long-read assembler. An example for Oxford Nanopore raw reads is:
flye \
–nano-raw reads.fastq.gz \
–genome-size 500m \
–out-dir flye_output \
–threads 8
- For Oxford Nanopore reads that have been processed to a higher-accuracy basecalled form, the appropriate Flye input mode should be selected according to the software version and documentation. PacBio HiFi data should be supplied using the HiFi-specific option rather than being treated as ordinary noisy long reads. The input mode must match the actual sequencing technology and read characteristics.
- An example of assembling PacBio HiFi reads using Hifiasm is:
hifiasm \
-o hifiasm_output \
-t 16 \
reads.hifi.fastq.gz
- Hifiasm is designed specifically for HiFi data and can produce primary and alternate assembly representations depending on the input and configuration. For diploid genomes, it may also support haplotype-resolved assembly. The output graph and sequence files should be examined to understand whether the resulting assembly represents primary contigs, alternate haplotypes, or phased sequences.
- Long-read assembly may require read correction, depending on the sequencing technology and assembler. Older long-read workflows often included a separate error-correction stage because raw reads could contain substantial sequencing errors. Correction methods compare overlapping reads and use redundant sequence information to improve base accuracy before assembly. Some modern assemblers perform correction internally, while others are designed to work directly with high-quality HiFi reads or basecalled nanopore data. Researchers should avoid unnecessary correction steps when the assembler already performs them or when they could introduce unwanted changes.
- Consensus polishing is another important stage in many long-read assembly workflows. Polishing improves the accuracy of assembled sequences by comparing the assembly with the original sequencing reads and correcting likely errors. Different tools are designed for different sequencing technologies. Racon can polish assemblies using long-read alignments, while Medaka is used for Oxford Nanopore consensus polishing with compatible models. PacBio workflows may use tools such as Arrow or the relevant PacBio consensus tools, depending on the data and software ecosystem. Short reads can also be used for polishing when suitable Illumina data are available.
- An illustrative polishing workflow using Racon for Oxford Nanopore reads is:
minimap2 \
-x map-ont \
assembly.fasta \
reads.fastq.gz \
> reads_to_assembly.paf
racon \
reads.fastq.gz \
reads_to_assembly.paf \
assembly.fasta \
> polished_assembly.fasta
- This example performs one Racon polishing round. Additional rounds may be useful in some workflows, but repeated polishing should not be applied automatically. The alignment and polishing results should be assessed to determine whether further correction improves accuracy. The appropriate minimap2 preset depends on the read type and assembly context.
- For Oxford Nanopore assemblies, Medaka can be used to generate a consensus sequence from reads aligned to an assembly. A simplified example is:
minimap2 \
-ax map-ont \
assembly.fasta \
reads.fastq.gz \
| samtools sort -o reads_to_assembly.bam
samtools index reads_to_assembly.bam
medaka_consensus \
-i reads.fastq.gz \
-d assembly.fasta \
-o medaka_output \
-t 8 \
-m MODEL_NAME
- MODEL_NAME is a placeholder for a model compatible with the sequencing chemistry, basecalling model, and Medaka version. It must be replaced with an appropriate supported model. Model selection is important because an incompatible model can reduce consensus accuracy. Researchers should consult the installed Medaka documentation and available model list rather than copying a placeholder literally.
- Long reads can also support scaffolding and chromosome-level genome assembly. Although long-read assemblers may generate highly contiguous contigs, additional technologies can help order and orient them into larger genomic structures. Hi-C sequencing captures information about physical proximity between DNA regions in the nucleus. These data can support chromosome-scale scaffolding by identifying genomic regions that frequently interact. Optical mapping and other long-range technologies can also provide structural information. Tools such as YaHS and 3D-DNA are used in certain Hi-C scaffolding workflows, while visualization and manual review may be performed using Juicebox.
- Haplotype resolution is an important advantage of some long-read assembly strategies. In diploid organisms, the two homologous chromosomes may contain different alleles and structural arrangements. An assembler that can distinguish these differences may generate separate haplotype assemblies rather than collapsing both chromosome copies into a single consensus. HiFi reads are particularly useful for this purpose because their combination of read length and accuracy can help distinguish closely related haplotypes. However, successful phasing depends on coverage, heterozygosity, genome complexity, and the assembly method.
- Long-read sequencing is especially useful for resolving repetitive regions. Repeats may be longer than conventional short reads, preventing reliable reconstruction of their genomic positions. A sufficiently long read can span an entire repeat and extend into unique sequence on both sides, providing evidence about its correct location. This can reduce repeat-associated assembly fragmentation and improve the reconstruction of complex genomic regions. However, repeats longer than the available reads, nearly identical duplicated regions, and highly complex rearrangements may remain difficult to resolve.
- Long-read assembly also presents specific challenges. Oxford Nanopore reads can contain systematic and random errors, depending on the sequencing chemistry, basecalling model, and data quality. PacBio HiFi reads generally have high consensus accuracy, but their read lengths and coverage may differ from those of nanopore data. Low coverage can result in fragmented assemblies, while contamination can introduce sequences from other organisms. High heterozygosity and polyploidy can create alternative graph paths, and duplicated regions may cause incorrect assembly connections. Long-read assembly therefore requires careful quality assessment, suitable algorithms, and independent validation.
- Metagenomic long-read assembly is particularly challenging because reads originate from multiple organisms with different abundances and genomic characteristics. Long reads may help connect genes and genomic regions within microbial communities, improving the reconstruction of metagenome-assembled genomes. However, closely related species and strains can share extensive sequence similarity, and uneven abundance can make genome reconstruction difficult. Long-read metagenomic assemblies should be evaluated using appropriate completeness, contamination, and taxonomic consistency measures.
- Assembly quality assessment should consider contiguity, completeness, accuracy, structural consistency, and biological plausibility. Common statistics include total assembly length, contig count, largest contig, N50, L50, and GC content. The N50 statistic measures contiguity, but a high N50 does not necessarily indicate a correct assembly. A misassembled sequence may be very long and still produce an impressive N50 value. Assembly statistics must therefore be interpreted alongside read mapping, structural validation, gene completeness, and contamination screening.
- QUAST is commonly used to evaluate assembly statistics and compare assemblies with suitable reference genomes. A basic example is:
quast.py \
assembly.fasta \
-o quast_output
- When a sufficiently close reference genome is available, it can be supplied for additional comparisons:
quast.py \
assembly.fasta \
-r reference.fasta \
-o quast_reference_output
- Reference-based comparisons can identify possible structural inconsistencies, but they must be interpreted carefully when the reference differs substantially from the organism being assembled. Genuine biological differences may otherwise be mistaken for assembly errors.
- BUSCO can be used to assess the completeness of conserved gene sets in an assembly. It identifies expected single-copy orthologous genes using an appropriate lineage dataset. This provides useful information about gene-space completeness, but it does not establish that the entire genome has been reconstructed correctly. An assembly can contain most expected genes while still having incorrect structural connections or unresolved repetitive regions.
- An example BUSCO command is:
busco \
-i assembly.fasta \
-m genome \
-l bacteria_odb12 \
-o busco_output \
-c 8
- The lineage dataset should be selected according to the organism and the datasets available in the installed BUSCO version. For larger eukaryotic genomes, an appropriate eukaryotic or more specific lineage dataset should be used instead.
- Read mapping is another important validation method. The original long reads can be aligned back to the assembled genome to examine coverage, alignment consistency, and possible structural errors. Minimap2 is widely used for long-read mapping. SAMtools can process the resulting alignment files and generate useful statistics.
- A simplified mapping workflow for Oxford Nanopore reads is:
minimap2 \
-ax map-ont \
assembly.fasta \
reads.fastq.gz \
| samtools sort -o mapped_reads.bam
samtools index mapped_reads.bam
samtools flagstat mapped_reads.bam
- For PacBio HiFi reads, the appropriate minimap2 preset should be used instead of the Oxford Nanopore preset. Read mapping can reveal regions with unusual coverage, inconsistent alignments, or possible structural problems. However, high mapping rates do not independently prove that an assembly is correct. Read alignments should be considered together with assembly graphs, coverage patterns, structural evidence, and independent sequencing data where available.
- Long-read genome assembly has numerous applications in microbial genomics, plant and animal genetics, human genomics, evolutionary biology, and comparative genomics. In microbial research, it can help reconstruct complete bacterial chromosomes and plasmids, identify antimicrobial resistance genes, and investigate genome organization. In plant and animal research, it supports the reconstruction of large and repetitive genomes, gene-family analysis, structural variant discovery, and the study of genome evolution. In human genomics, long-read assemblies can help characterize complex genomic regions, repeat expansions, structural variants, and haplotype-specific sequences.
- Long-read sequencing is also valuable in transcriptomics and epigenomics, although these applications require workflows designed for RNA or DNA modification analysis rather than conventional genome assembly alone. Long-read RNA sequencing can help reconstruct full-length transcript isoforms, while nanopore signal analysis and other methods can provide evidence about certain nucleotide modifications. These analyses should be distinguished from genomic DNA assembly because their objectives and data-processing requirements differ.
- Hybrid assembly combines long-read and short-read sequencing data to use the strengths of both technologies. Long reads provide the continuity needed to span repeats and connect distant genomic regions, while accurate short reads can contribute to base-level correction and validation. Some hybrid assemblers use both data types during assembly, whereas other workflows first assemble long reads and then polish the resulting sequences with short reads. The appropriate strategy depends on the genome, sequencing quality, and intended level of completeness.
- For reproducible long-read assembly, researchers should preserve the original sequencing files, including raw signal data when available, basecalled reads, and intermediate files. They should document the sequencing platform, chemistry, library preparation, read length distribution, basecalling software and model, sequencing coverage, and quality-control results. Assembly software versions, parameters, polishing rounds, reference data, and validation methods should also be recorded. Checksums can help verify file integrity, while workflow managers and containerized environments can improve reproducibility across computational systems.
- Long-read genome assembly has significantly improved the ability of researchers to reconstruct complex genomes and investigate genomic variation. PacBio HiFi and Oxford Nanopore sequencing provide complementary approaches for generating long DNA reads, while assemblers such as Canu, Flye, Hifiasm, and Shasta provide computational methods for reconstructing genomic sequences. The success of long-read assembly depends on read accuracy, read length, coverage, genome complexity, and appropriate computational methods. Careful polishing, quality assessment, and validation remain essential because high contiguity alone does not guarantee a complete or accurate genome. Long-read assembly is therefore a major component of modern genome analysis and an important foundation for more complete and reliable genomic research.