{
  "id": "long-read-sequencing",
  "title": "Long-Read Sequencing",
  "lang": "en",
  "source_lang": "en",
  "summary": "Long-read sequencing determines extended DNA or RNA sequences from individual molecules, revealing genome structure, genetic variation, transcript diversity, and molecular modifications.",
  "category": "science",
  "subcategory": "biology",
  "type": "concept",
  "aliases": [
    "Long-read DNA sequencing",
    "third-generation sequencing"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/long-read-sequencing/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/long-read-sequencing/",
      "json": "https://aiwiki.page/zh/long-read-sequencing.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Long-read sequencing is a group of DNA sequencing technologies that determine the order of bases across individual nucleic-acid molecules thousands to hundreds of thousands of bases long. Some methods also sequence RNA directly. Unlike short-read methods, which generally produce reads a few hundred bases long or less, long-read methods preserve extensive information about which sequence features occur together on the same molecule. This makes them particularly useful for reconstructing genomes, resolving repetitive regions, identifying large genetic changes, and distinguishing closely related gene copies.\n\nThe two principal technological approaches are single-molecule real-time sequencing, associated with Pacific Biosciences, and nanopore sequencing, associated with Oxford Nanopore Technologies. They differ in their physical measurements, error patterns, read-length distributions, and ability to detect molecular modifications. Their development has helped transform genome analysis from the study of relatively accessible regions into the reconstruction of entire chromosomes and the comparison of complete, individually resolved genomes."
    },
    {
      "heading": "Read length, information content, and technological development",
      "text": "A sequencing read is an inferred sequence obtained from a nucleic-acid molecule or a defined portion of one. There is no universal minimum length that makes a read “long,” but the term usually refers to reads extending for several kilobases or more. High-accuracy long reads commonly span approximately 10–25 kilobases, while nanopore datasets can contain reads tens or hundreds of kilobases long. Exceptionally long nanopore reads have exceeded one megabase. Such records demonstrate technical capability rather than the typical composition of a sequencing run.\n\nThe importance of length arises from the organization of genomes. A repeat longer than a short read may appear indistinguishable from several other copies elsewhere in a chromosome. A longer read can cross the repeat and reach distinctive sequence on both sides, establishing its genomic position. Similarly, one molecule can connect multiple variants, reveal an insertion together with its flanking sequence, or traverse several exons of a transcript. Length therefore supplies linkage information, not merely more bases per observation. Nevertheless, a repeat longer than the available reads, especially one with nearly identical copies, may remain unresolved.\n\nSeveral metrics describe a dataset. Mean and median read length summarize its distribution, whereas read N50 is the length at which reads of that length or longer contain at least half of all sequenced bases. N50 is weighted by bases rather than by the number of reads. Total sequence yield and genome coverage are separate quantities. For a genome of length \\(G\\), nominal coverage is\n\n\\[C=\\frac{\\sum_i L_i}{G},\\]\n\nwhere \\(L_i\\) is the length of read \\(i\\). This average does not imply uniform coverage, independent observations, or equal representation of both chromosome copies.\n\nLong-read sequencing emerged from efforts to measure individual molecules without requiring large populations of synchronized copies. Pacific Biosciences introduced its first commercial instrument in 2011. Oxford Nanopore’s MinION became available through an early-access program in 2014. Early implementations had substantially higher per-read error rates than established short-read sequencing, encouraging hybrid workflows that used short reads for correction and long reads for structural continuity.\n\nSubsequent improvements changed that balance. Circular consensus sequencing produced highly accurate long reads, while nanopore improvements included revised pores, motor proteins, signal-processing methods, and complementary-strand sequencing. “Third-generation sequencing” is often used as a synonym, although it is an imprecise historical classification. The distinction that matters operationally is how a platform measures molecules and what information survives library preparation, rather than which technological “generation” it is assigned to."
    },
    {
      "heading": "Single-molecule real-time sequencing and circular consensus",
      "text": "Single-molecule real-time sequencing observes DNA synthesis by an individual DNA polymerase. Its central optical structure is a zero-mode waveguide: a nanoscale aperture that confines the illuminated observation volume near its base. This small volume makes it possible to distinguish signals from nucleotides held by a polymerase from the background fluorescence of freely diffusing nucleotides.\n\nThe four nucleotide types carry distinguishable fluorescent labels attached to their phosphate groups. As a polymerase incorporates a nucleotide into the growing strand, the label remains near the active site long enough to generate a detectable pulse. Incorporation releases the dye-bearing phosphate group, leaving the newly synthesized DNA without that fluorescent label. The sequence of pulses is converted into a sequence of bases. Thousands or millions of observation sites can operate in parallel, depending on the instrument.\n\nLibrary molecules typically use hairpin adapters to connect the two strands of a double-stranded DNA fragment into a closed template called a SMRTbell. The polymerase can proceed around this template, reading one insert strand, an adapter, the complementary strand, and the other adapter. The resulting continuous polymerase read may therefore contain several observations of the same original insert. Those observations, separated computationally into subreads, are not independent DNA molecules.\n\nCircular consensus sequencing combines repeated observations of one template to infer a more accurate sequence. Pacific Biosciences uses the term HiFi for high-accuracy circular consensus reads, conventionally associated with predicted accuracy of at least 99%. Well-supported reads often exceed that threshold. Consensus is effective because many errors in individual passes do not recur at exactly the same positions. Sequence context and other systematic effects can still limit accuracy, so additional passes do not eliminate every error.\n\nInsert length and consensus accuracy are linked. For a finite polymerase observation time, a shorter insert can be traversed more often than a longer one. Longer inserts provide more genomic continuity but may receive fewer complete passes. Library design and analysis thresholds therefore balance molecule length, productive observations, and the accuracy needed for the intended application.\n\nThe instrument also records polymerase kinetics, including intervals between incorporation events. Some chemical modifications alter these timing patterns, allowing selected DNA modifications to be inferred from native templates. This does not make every modified base equally detectable: performance depends on the modification, sequence context, chemistry, and analytical model. Similarly, high consensus accuracy should not be confused with complete genome recovery. Missing molecules, repetitive sequences, uneven coverage, and library biases can remain important even when the reads themselves are highly accurate."
    },
    {
      "heading": "Nanopore sequencing and electrical signal interpretation",
      "text": "Nanopore sequencing measures changes in electric current as a nucleic-acid strand moves through a nanoscale pore. In commercial biological-pore systems, the pore is embedded in a membrane separating electrolyte-filled compartments. An applied electrical potential drives an ionic current through it. DNA or RNA occupying the pore changes that current in a sequence-dependent manner.\n\nA motor protein controls the movement of the strand, slowing translocation sufficiently for measurement. The electrical signal generally reflects several neighboring bases within the pore’s sensing region rather than one isolated base at a time. Consequently, sequence inference is a context-dependent signal-decoding problem. A base-calling program converts the time series into bases, commonly using models based on artificial neural networks. Different models can trade computational speed against accuracy.\n\nUnlike synthesis-based sequencing, nanopore sequencing does not require the measured strand to be copied during signal acquisition. Its read length is therefore strongly influenced by the physical integrity of the input molecule, successful adapter attachment, motor performance, and uninterrupted passage through the pore. Routine libraries often yield reads tens of kilobases long; specialized ultra-long workflows preserve much larger fragments. A single very long read can bridge a difficult region, but a useful assembly usually requires multiple supporting molecules.\n\nNanopore errors are not distributed uniformly. Homopolymers, certain repeated motifs, and some modified-base contexts have posed particular difficulties because similar sequences can produce overlapping signal patterns. Improvements in pore design and signal analysis have reduced these problems, but performance must be evaluated for the actual chemistry and software used. An accuracy figure from one platform generation cannot automatically be applied to another.\n\nDuplex sequencing identifies and combines observations from both complementary strands of the same DNA molecule. Agreement between the two measurements can improve accuracy, although not every molecule yields a usable duplex pair. A distinction is therefore necessary between simplex accuracy, duplex accuracy, and consensus accuracy obtained from multiple independent molecules covering a genomic position.\n\nNanopore sequencing also allows data-dependent decisions during a run. In adaptive sampling, an initial segment of electrical signal or inferred sequence is compared with a target definition. A molecule judged unwanted can be ejected by reversing the voltage, freeing the pore for another molecule. This provides software-controlled enrichment without capturing each target physically, although its effectiveness depends on target size, fragment length, decision speed, and sample composition.\n\nThe same general sensing principle supports direct RNA sequencing and modification analysis. It also creates a substantial computational requirement: raw electrical signals are richer and larger than the final base sequences, and reanalysis with improved models can change both sequence and modification calls."
    },
    {
      "heading": "Sample integrity and library preparation",
      "text": "The practical upper limit on read length is often established before sequencing begins. Long DNA molecules can break during tissue disruption, mixing, pipetting, purification, storage, or repeated freezing and thawing. Their preservation requires extraction methods that release DNA while limiting mechanical shear and chemical damage. The relevant objective is not simply high DNA concentration, but a sufficiently large population of intact, usable molecules.\n\nSample type strongly affects that objective. Fresh tissue, cultured cells, and some blood-derived preparations can yield high-molecular-weight DNA. Formalin-fixed material often contains fragmentation and crosslinks, while ancient DNA is usually too fragmented to exploit the defining length advantage of these technologies. Plant tissues may contain polysaccharides and phenolic compounds; environmental extracts may contain humic substances. Such contaminants can inhibit enzymatic preparation steps or interfere with sequencing.\n\nLibrary preparation adds the molecular structures required by the instrument. Ligation-based nanopore libraries attach adapters associated with a motor protein, using DNA ligase to form covalent joins. Some rapid preparations use transposase-mediated adapter insertion, which simplifies handling but also introduces breaks. Circular-consensus libraries require construction of closed templates from suitably prepared double-stranded fragments. Platform-specific cleanup and size-selection steps influence which molecules reach the instrument.\n\nSize selection can remove short fragments that would otherwise consume sequencing capacity. It can also reduce total yield or preferentially exclude biologically important small molecules. A workflow optimized for long human chromosomal DNA may recover small plasmids poorly. In mixed microbial samples, extraction and selection can distort the relative abundance of organisms whose cells differ in ease of lysis or whose genomes differ in molecular form.\n\nMany long-read workflows avoid amplification. This helps retain native base modifications and reduces biases associated with polymerase chain reaction. Amplification remains useful when input is limited or a specific region is targeted, but it creates copies rather than additional independent source molecules. It can introduce chimeras, alter allele balance, and remove the native modification patterns that direct sequencing could otherwise detect.\n\nQuality control therefore includes complementary measurements. DNA mass does not establish fragment length, and absorbance ratios do not guarantee freedom from inhibitors. Electrophoretic methods and related instruments estimate size distributions, while trial sequencing can reveal whether apparently suitable material produces productive reads. The best preparation is application-dependent: preserving ultra-long molecules, obtaining uniform high-accuracy coverage, recovering all members of a microbial community, and processing low-input clinical specimens are different technical goals."
    },
    {
      "heading": "Genome assembly, haplotypes, and complete chromosomes",
      "text": "Genome assembly reconstructs larger sequences from overlapping reads. Long reads reduce ambiguity because individual molecules can extend through repetitive elements into distinctive flanking sequence. Assemblers use sequence overlaps or graph representations to determine which observations are compatible and to build contiguous sequences, or contigs. Accurate long reads reduce the need for extensive preliminary correction, but assembly remains a process of inference rather than direct concatenation.\n\nA crucial difficulty is that most human cells contain two homologous copies of each autosome. These copies are similar but not identical. An assembler may collapse them into one sequence, separate them into two, or switch between them along a contig. Haplotype phasing assigns variants and sequence segments to the same inherited chromosome copy. Long reads provide direct phase information when multiple distinguishing variants occur on one molecule, while parental sequence data and chromosome-conformation measurements can extend the connections over longer distances.\n\nRepeats remain challenging even with long reads. Recently duplicated regions may be almost identical over hundreds of kilobases, and centromeres contain large arrays of satellite repeats. Ultra-long reads can bridge some structures, while high-accuracy reads distinguish small differences among repeat copies. Combining the two types has been especially productive because physical span and per-base discrimination solve different parts of the problem.\n\nIn 2022, the Telomere-to-Telomere Consortium published the T2T-CHM13 assembly, adding nearly 200 million bases absent from the earlier human reference framework and resolving many previously unfinished regions. CHM13 was derived from a nearly homozygous cell line, which reduced the difficulty of separating divergent homologues. The assembly did not include a Y chromosome; a complete human Y-chromosome sequence from a different source was reported in 2023. These achievements were major demonstrations of complementary long-read technologies, not evidence that every human genome had become straightforward to assemble.\n\nA complete sequence from one source cannot represent all human genetic diversity. The Human Pangenome Reference Consortium’s 2023 draft incorporated assemblies from 47 individuals, illustrating the move toward multiple reference haplotypes. A pangenome represents shared and variable sequence across genomes, potentially reducing the bias introduced by mapping every sample against a single linear reference.\n\nAssembly quality requires several independent checks. A high contig N50 measures continuity but cannot establish correctness. Evaluation also examines base accuracy, missing sequence, duplicated regions, chromosome-scale organization, and phase switches. Short sequence patterns, independent reads, genetic maps, or other long-range measurements can expose errors. A chromosome-length contig can still contain an incorrect join, and a gap-free assembly can still misrepresent the order or number of repeat units."
    },
    {
      "heading": "Structural variation and human genetic analysis",
      "text": "Long reads are particularly informative for structural variation, commonly defined as genomic differences involving approximately 50 bases or more. This category includes deletions, insertions, inversions, duplications, translocations, and more complex rearrangements. Some alter the amount of DNA; others change its orientation or chromosomal location without a corresponding change in total copy number.\n\nShort reads often reveal a rearrangement indirectly through abnormal read-pair spacing, split alignments, or coverage changes. A long read may traverse the breakpoint and show the altered junction directly. For an insertion, it may recover both the inserted sequence and the flanking reference sequence. Multiple independent reads are still important, because individual molecules can be damaged or chimeric and alignments can be ambiguous in repetitive regions.\n\nRepeat expansions are another important application. Long reads can measure the length and internal sequence of a tandem-repeat array, identify interruptions within it, and connect it to flanking variants. Examples include expansions in FMR1 and C9orf72. Such loci can be difficult for amplification-based assays or conventional short-read analysis, especially when expansions are large or compositionally biased. The ability to read a repeat does not by itself establish clinical interpretation: the relevant repeat structure, inheritance pattern, and relationship to disease must be evaluated separately.\n\nPhasing adds a different type of information. Two variants in one gene may lie on the same inherited copy, in cis, or on different copies, in trans. This distinction can affect the interpretation of recessive genetic conditions. Long molecules can connect the variants directly when their separation is within the molecule’s span. In highly similar gene families, they can also link a candidate variant to sequence that distinguishes a functional gene from a related pseudogene.\n\nCancer analysis introduces further complexity. Tumors can contain mixed cell populations, abnormal chromosome numbers, subclonal rearrangements, and extensive copy-number changes. Long reads can clarify complex junctions and rearrangement structures, but sensitivity to a rare event depends on the number of independent molecules sampled from the relevant cell population. A very long read does not compensate for insufficient sampling of a low-frequency variant.\n\nClinical implementation consequently requires validation by variant class and genomic context. Sensitivity for single-base substitutions cannot stand in for sensitivity to expansions or rearrangements. Reference materials, reproducibility studies, controls, and appropriate confirmatory methods establish what an assay can support. Sequencing identifies molecular observations; diagnosis additionally depends on phenotype, inheritance, and established clinical interpretation. Accurate sequencing also does not make every observed variant clinically meaningful, since many remain of uncertain significance."
    },
    {
      "heading": "Transcript isoforms and epigenetic information",
      "text": "Long-read RNA sequencing addresses a central limitation of fragmented transcript measurements: short pieces from the same gene do not always reveal which exons occurred together in the original molecule. Through alternative splicing, one gene can produce multiple transcripts with different combinations of exons. A read spanning most or all of a transcript can connect splice junctions, transcriptional start regions, and alternative terminal regions within one observation.\n\nTwo broad approaches are used. Complementary-DNA sequencing first reverse-transcribes RNA into DNA and sequences the resulting molecules, sometimes after amplification. Direct RNA sequencing measures RNA itself as it passes through a nanopore. The former can produce accurate transcript sequences and works with a range of library strategies; the latter preserves information carried by the native RNA molecule but has different input, throughput, and error characteristics.\n\nIn a common direct-RNA workflow, an adapter is attached at the polyadenylated end and the RNA is read from the 3′ end toward the 5′ end. Reverse transcription may be included to improve sequencing behavior even though the RNA strand generates the measured signal. Incomplete translocation or molecule breakage can leave the 5′ end underrepresented. Consequently, “long-read transcriptomics” does not imply that every read corresponds to a complete, intact messenger RNA.\n\nTranscript identification and transcript quantification are distinct problems. A long read may establish an exon combination convincingly, but accurate abundance estimates still require adequate sampling and control of length-dependent recovery, amplification, and mapping biases. Apparent new isoforms can result from reverse-transcription artifacts, chimeric molecules, or alignment mistakes near splice junctions. Support from multiple molecules and compatible expression evidence helps distinguish biological transcripts from technical artifacts.\n\nNative-molecule sequencing also connects sequence to epigenetic information. DNA methylation can change polymerase kinetics or nanopore current patterns, depending on the platform. Nanopore models can infer selected modified bases directly from electrical signals, while real-time synthesis measurements infer modifications through their effects on incorporation behavior. Neither method requires the same chemical conversion used in conventional bisulfite sequencing.\n\nBecause a long read contains both sequence variants and modification measurements, methylation can be assigned to a particular haplotype. This is useful for studying allele-specific regulation and genomic imprinting, in which parental origin influences gene activity. In microbial genomes, characteristic methylation patterns can also reflect restriction–modification systems.\n\nRNA modifications create a more difficult inference problem because chemical modifications, sequence context, RNA structure, and measurement noise can produce related signal changes. Detection models therefore need suitable controls and modification-specific validation. An altered signal is evidence requiring interpretation, not an automatic identification of a particular chemical group or a direct measurement of its biological effect."
    },
    {
      "heading": "Microbial communities, comparative genomics, and biodiversity",
      "text": "In metagenomics, DNA is sequenced from a community rather than from a single isolated organism. Long reads can connect microbial genes with neighboring sequence, distinguish closely related strains, and improve recovery of complete or nearly complete microbial genomes. These advantages are particularly important where repeated sequences or closely related organisms fragment assemblies based on short reads.\n\nPhysical linkage changes what can be inferred from a community sample. Detecting an antibiotic-resistance gene establishes its presence but does not necessarily identify the organism carrying it. A long read spanning that gene and characteristic flanking sequence may connect it to a host chromosome or plasmid. This can clarify the genomic context of horizontal gene transfer. Nevertheless, plasmid host assignment remains uncertain when the plasmid is sequenced as a separate molecule without a direct host-specific connection.\n\nLong reads can also improve the reconstruction of viral genomes, biosynthetic gene clusters, and repeated ribosomal operons. Some microbial chromosomes can be assembled as single circular sequences when coverage and read quality are sufficient. A computationally circularized contig is not, by itself, proof that every base is correct or that the source population contains only one strain. Closely related strains can be merged or misassembled if their differences are not adequately resolved.\n\nCommunity sampling creates a major coverage imbalance. Abundant organisms may receive deep coverage while rare organisms remain represented by only a few molecules. Increasing total yield helps, but extraction biases, genome size, and strain diversity also influence recovery. Amplified long marker genes, including full-length 16S ribosomal RNA genes, provide another strategy for profiling communities. They are targeted surveys rather than unbiased measurements of all community DNA, and taxonomic resolution depends on the marker and reference database.\n\nComparative genomics benefits from a similar gain in structural context. Long-read assemblies can expose duplicated genes, rearrangements, mobile elements, and lineage-specific sequence that fragmented references miss. These features contribute to studies of evolution, adaptation, and the organization of complex gene families. Better assembly can also prevent apparent gene loss from being mistaken for biological absence when the sequence was merely missing from an earlier reference.\n\nIn biodiversity research, long reads have supported reference-genome projects across animals, plants, fungi, and other organisms. Polyploid plants and highly heterozygous species remain difficult because multiple related chromosome copies must be separated accurately. Large genomes with abundant transposable elements require substantial sequencing and computational resources.\n\nPortable nanopore instruments additionally allow sequencing near collection sites. Portability shortens some logistical chains, but it does not remove the need for suitable extraction, contamination controls, reagents, power, and analysis. Environmental detection also requires careful interpretation: a recovered sequence can establish the presence of genetic material without establishing organism viability or ecological activity."
    },
    {
      "heading": "Accuracy, computational demands, and standards of evidence",
      "text": "Long-read performance cannot be reduced to one accuracy number. Raw-read accuracy concerns individual observations, consensus accuracy concerns a sequence inferred from repeated observations, and assembly accuracy concerns reconstructed genomic sequence. Variant-calling accuracy is another level, evaluated through sensitivity, precision, and the types of events tested. Strong performance at one level does not guarantee equal performance at the others.\n\nBase-quality scores commonly use a logarithmic error scale:\n\n\\[Q=-10\\log_{10}(p),\\]\n\nwhere \\(p\\) is the estimated probability of an incorrect base call. A score of Q20 corresponds to an estimated error probability of 1%, and Q30 to 0.1%. These are model-based estimates rather than direct guarantees. Their usefulness depends on calibration: calls assigned a particular quality should show approximately the corresponding empirical error frequency in appropriate benchmark data.\n\nRead length and coverage also interact with accuracy. Greater coverage can reduce random consensus errors, but systematic errors may persist if many observations share the same context-dependent failure. Longer reads improve mapping across repeats but may require more computation and can still be misplaced when reference sequences are incomplete. A benchmark that excludes difficult repetitive regions can therefore overstate performance for precisely the regions that motivate long-read sequencing.\n\nComputational workflows include base calling, filtering, mapping or assembly, phasing, variant detection, and annotation. Some steps use graphics processing units, while assembly and signal retention can demand substantial memory and storage. Different analyses require different retained data: base sequences may suffice for one task, whereas improved modification calling or re-base-calling may require the original signals or platform-specific intermediate measurements.\n\nReproducibility depends on recording more than the instrument name. Pore or sequencing chemistry, library kit, base-caller version, model, filtering thresholds, reference assembly, and variant-calling software can all affect results. Rapid improvements create a moving target: published error rates or throughput figures describe particular configurations, not permanent properties of “long reads.”\n\nReference materials and benchmark regions provide a basis for comparison, but their limits matter. Well-characterized human samples cover only part of human diversity, and reliable benchmark calls remain harder to establish in highly repetitive, structurally variable, or mosaic regions. Orthogonal evidence is valuable because two methods with different failure modes can test the same interpretation more strongly than repeated analysis through one pipeline.\n\nLong-read data also raise data-privacy issues. They can expose inherited variants, chromosome-scale haplotypes, and modification patterns, while sample metadata may add identifying context. Controlled access, consent terms, and rules for data sharing apply to both processed sequences and the richer underlying measurements. The scientific value of retaining raw data must therefore be considered alongside the governance of information derived from identifiable biological samples."
    }
  ],
  "links": [
    "dna-sequencing",
    "rna",
    "genome",
    "chromosome",
    "dna-polymerase",
    "nucleotide",
    "dna",
    "nanopore-sequencing",
    "electric-current",
    "protein",
    "artificial-neural-network",
    "ancient-dna",
    "dna-ligase",
    "plasmid",
    "polymerase-chain-reaction",
    "gel-electrophoresis",
    "genome-assembly",
    "centromere",
    "pangenome",
    "gene",
    "diagnosis",
    "rna-sequencing",
    "alternative-splicing",
    "messenger-rna",
    "epigenetics",
    "dna-methylation",
    "metagenomics",
    "horizontal-gene-transfer",
    "evolution",
    "biodiversity",
    "transposable-element",
    "calibration",
    "graphics-processing-unit",
    "data-privacy"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}