# [Oxford Nanopore Sequencing](/knowledge/bioinformatics/oxford-nanopore-sequencing-from-sample-to-base-calls) Accuracy: Mechanisms and Optimization

## Introduction to Oxford [Nanopore Sequencing](/blog/guides/nanopore-sequencing) Accuracy

Oxford Nanopore Technologies (ONT) sequencing has transformed genomics by enabling real-time, long-read, and portable DNA/RNA sequencing. Unlike sequencing-by-synthesis methods that read nucleotides through enzymatic incorporation and optical detection, [nanopore sequencing](/blog/guides/nanopore-sequencing) threads a single DNA or RNA molecule through a protein pore embedded in an electrically resistant membrane. As the molecule translocates, it modulates an ionic current, and the resulting signal is decoded into a [nucleotide sequence](/knowledge/molecular-biology/nucleotide-sequence). The central challenge of this technology has always been accuracy: translating a continuous, analog electrical signal into a discrete digital sequence with minimal error.

The field has progressed rapidly. Early MinION releases in 2014–2015 produced raw read accuracies of approximately 60–70%, making single-read analysis unreliable for most applications. By 2018, with the introduction of Guppy basecalling and improved pore chemistry, raw read accuracy reached 85–90%. The R9.4.1 flow cell and its associated chemistry pushed this to 92–95% for typical libraries. The R10.4.1 pore, released in 2021, improved homopolymer resolution and brought raw read accuracy to 95–98% for many datasets. Consensus accuracy—derived from multiple reads covering the same region—can now exceed Q50 (99.999%) for small genomes, rivaling short-read platforms.

Why does this matter? Accuracy determines the ceiling for every downstream application. Variant calling, [structural variant](/blog/guides/structural-variant) detection, methylation profiling, and clinical diagnostics all depend on the fidelity of the underlying sequence. Understanding the mechanistic sources of error, the algorithmic tools that mitigate them, and the experimental strategies that maximize accuracy is essential for any researcher designing a nanopore experiment.

### What is Oxford Nanopore Sequencing?

Nanopore sequencing is a single-molecule, real-time technology. A DNA library is prepared with adapter sequences that load onto a motor protein, typically a helicase or translocase. This protein unwinds the double-stranded DNA and feeds a single strand through a nanopore—a modified protein, usually MspA or CsgG, embedded in a synthetic polymer membrane. The membrane separates two chambers filled with ionic buffer. A voltage bias (typically −180 mV) drives ions through the pore, creating a baseline current of approximately 100–200 pA. When a nucleotide enters the pore, it partially blocks the current. The magnitude of the block depends on the combination of nucleotides occupying the pore's sensing region—typically 4 to 5 nucleotides at a time for R9 pores, and 9 to 10 for R10 pores.

The current is sampled at 4 kHz (or higher on newer devices) and recorded as raw signal. This signal is then processed by a basecaller, which converts the current traces into a sequence of A, C, G, and T. The process is continuous: as the motor protein ratchets the DNA through the pore, the current changes in a predictable but complex manner. The relationship between the [nucleotide sequence](/knowledge/molecular-biology/nucleotide-sequence) and the current is not one-to-one; it is a function of the k-mer (the short sequence of nucleotides in the sensing region) and the surrounding context. This is why basecalling is fundamentally a pattern recognition problem.

### Accuracy Metrics: Read-level vs Consensus Accuracy

Accuracy is reported in two distinct ways, and conflating them is a common source of confusion.

**Read-level (raw read) accuracy** refers to the accuracy of a single, individual sequencing read. It is typically measured by aligning the read to a reference genome and calculating the percentage of correctly identified bases. For a 10 kb read with 95% accuracy, approximately 500 bases are incorrect—a mix of substitutions, insertions, and deletions. Read-level accuracy is the relevant metric for applications that rely on individual molecules, such as detecting base modifications in a single DNA strand or analyzing RNA isoforms without consensus support.

**Consensus accuracy** refers to the accuracy of a sequence derived from multiple reads aligned to the same region. Because errors in nanopore sequencing are largely stochastic (though some are systematic), aligning 20–50 reads and calling a consensus can reduce error rates dramatically. For a small genome like *E. coli* (4.6 Mb), a typical R10.4.1 run with 50× coverage can produce a consensus accuracy exceeding Q50 (99.999%). For human-scale genomes, consensus accuracy of Q40 (99.99%) is achievable with sufficient coverage. Consensus accuracy is the metric that matters for de novo assembly, variant calling, and most clinical applications.

The distinction is critical: a platform with 95% raw read accuracy can still produce reference-quality assemblies if the error profile is random and coverage is sufficient. Conversely, systematic errors—those that occur at the same position in every read—cannot be corrected by depth alone.

## Mechanisms of Error in Nanopore Sequencing

Understanding error mechanisms requires a detailed look at the physics and chemistry of the sequencing process. Errors arise from three interconnected sources: the translocation dynamics of the DNA through the pore, the electrical properties of the pore itself, and the computational challenge of decoding the resulting signal.

### Ionic Current and Base Calling

The nanopore sensing region contains a narrow constriction where the DNA backbone interacts with the pore wall. As each nucleotide passes through, it alters the ionic current in a way that depends on the identity of the nucleotide and its neighbors. For R9.4 pores, the sensing region spans approximately 4 nucleotides (a 4-mer). For R10.4 pores, the sensing region is longer—approximately 9–10 nucleotides—which provides more information per position and improves discrimination, particularly for homopolymers.

The current signal is not a simple step function. As the motor protein ratchets the DNA in discrete steps (typically 1 nucleotide per step for R10 pores, but variable for R9), the current transitions between states. These transitions are not instantaneous; they have a characteristic dwell time and a transient current spike. The basecaller must segment the continuous signal into discrete events, each corresponding to a nucleotide translocation step, and then assign a k-mer to each event. Errors in segmentation—merging two events or splitting one—produce insertions and deletions in the final sequence.

The electrical signal is also subject to noise. Thermal fluctuations, pore blockages by contaminants, and variations in the membrane's electrical properties all contribute to baseline drift and transient spikes. The signal-to-noise ratio is a fundamental limit on accuracy. R10 pores were designed to increase this ratio by extending the sensing region, which reduces the degeneracy of the current signal for different k-mers.

### Sources of Systematic Errors

While many errors are stochastic, several are systematic and reproducible across runs.

**Homopolymer errors** are the most notorious. When a sequence contains a run of identical nucleotides (e.g., AAAAA), the motor protein translocates each nucleotide through the pore, but the current signal for each position is nearly identical. The basecaller must count the number of steps to determine the homopolymer length. For R9 pores, this counting is error-prone, leading to frequent under- or over-counting. R10 pores, with their longer sensing region, provide a more distinct signal for each position within a homopolymer, reducing but not eliminating this error. Homopolymer errors are primarily deletions or insertions, and they are the dominant error type in R9 data.

**Methylation and base modifications** alter the current signal. 5-methylcytosine (5mC) and N6-methyladenine (6mA) produce current signatures that differ from their unmodified counterparts. While this enables direct epigenetic detection, it also confuses basecallers that are not explicitly trained to recognize modified bases. A basecaller trained on unmodified DNA may mis-call a methylated cytosine as thymine or produce a deletion at that position. The R10.4.1 pore, combined with basecallers that include modified-base models, can distinguish 5mC from cytosine with high accuracy, but this requires the appropriate model and sufficient coverage.

**Sequence context effects** arise from the fact that the current for a given k-mer depends on the surrounding sequence beyond the immediate sensing region. This is particularly problematic for GC-rich regions, where the DNA secondary structure can slow translocation or cause the motor protein to pause. Pauses produce long dwell times that the basecaller may misinterpret as multiple events, generating spurious insertions.

**Motor protein stalling and reversal** occurs when the helicase encounters a lesion, a secondary structure, or a modified base that it cannot process. The motor may pause, reverse, or dissociate entirely. Reversal produces a "double-read" of the same sequence, which the basecaller may interpret as a duplication. Dissociation ends the read prematurely. These events are more common in damaged or highly modified DNA and contribute to both reduced throughput and increased error rates.

**Pore damage and contamination** are run-level factors. Over time, pores can become blocked by DNA fragments, proteins, or buffer contaminants. A blocked pore produces a constant, non-informative current. Some blockages are reversible with a "pore scrub" protocol, but repeated scrubbing can damage the pore. Damaged pores produce noisy or biased signals, and reads from these pores should be filtered out during downstream analysis.

## Basecalling Algorithms and Their Impact on Accuracy

Basecalling is the computational process of converting raw current signals into nucleotide sequences. The choice of basecaller and its configuration has a profound impact on accuracy, often more than the choice of flow cell or chemistry.

### Hidden Markov Models vs Neural Networks

Early basecallers, such as Metrichor and the first versions of Guppy, used hidden Markov models (HMMs) to decode the signal. An HMM treats the underlying sequence as a hidden state, with the observed current as an emission. The model learns the expected current for each k-mer and uses the Viterbi algorithm to find the most likely sequence given the observed signal. HMMs are computationally efficient and interpretable, but they struggle with the complex, non-linear relationships between sequence and current. They also assume that each event corresponds to a single nucleotide step, which is not always true.

Modern basecallers, including Guppy (since version 3.0) and Bonito, use deep neural networks—specifically, recurrent neural networks (RNNs) with long short-term memory (LSTM) units or transformer architectures. These models learn the mapping from raw signal to sequence directly, without explicit event segmentation. They can capture long-range dependencies in the signal, handle variable dwell times, and model the transition dynamics between k-mer states. The result is a significant improvement in accuracy, particularly for homopolymers and complex sequence contexts.

The architecture matters. Guppy uses a convolutional front-end followed by an LSTM and a connectionist temporal classification (CTC) decoder. Bonito, ONT's research basecaller, uses a similar architecture but is trained on larger and more diverse datasets. Third-party basecallers, such as Megalodon and Dorado (ONT's current production basecaller), offer additional features like modified-base calling and methylation detection. The choice of basecaller is not just about accuracy; it also affects speed, GPU requirements, and the ability to call modified bases.

### Training Data and Model Updates

The accuracy of a neural network basecaller is determined by its training data. ONT trains its production models on a diverse set of genomes, including human, bacterial, and plant DNA, with known reference sequences. The training data includes a wide range of GC content, repeat structures, and base modifications. Models are updated regularly—often several times per year—and each update typically brings a small but measurable accuracy improvement.

For researchers, the practical implication is that basecalling should be performed with the latest model available for the specific pore and chemistry used. Using an outdated model can reduce accuracy by several percentage points. Additionally, some models are specialized: a "high-accuracy" model may be slower but more accurate, while a "fast" model trades accuracy for speed. For most applications, the high-accuracy model is the right choice, even if it requires a GPU and longer processing time.

The basecalling quality score (Q-score) is a per-base estimate of accuracy, analogous to the Phred score used in short-read sequencing. A Q-score of 20 corresponds to 99% accuracy, Q30 to 99.9%, and Q40 to 99.99%. These scores are produced by the basecaller and can be used to filter reads. However, Q-scores are not perfectly calibrated; they tend to overestimate accuracy for homopolymers and underestimate it for complex regions. It is prudent to treat Q-scores as relative indicators rather than absolute guarantees.

## Library Preparation and Chemistry Effects on Accuracy

The accuracy of nanopore sequencing is not solely determined by the pore and basecaller. Library preparation choices—the enzymes, adapters, and amplification steps—directly influence the quality of the DNA that enters the pore and, consequently, the error rate.

### Ligation vs Rapid Kits

ONT offers two main library preparation strategies: ligation-based and rapid (tagmentation-based) kits.

**Ligation kits** (e.g., SQK-LSK114) involve end-repair and dA-tailing of the DNA, followed by ligation of sequencing adapters. The end-repair step uses a combination of T4 DNA polymerase and T4 polynucleotide kinase to create blunt ends and phosphorylate the 5' ends. The dA-tailing step uses Taq polymerase to add a single adenine to the 3' ends. The adapters, which contain a complementary thymine overhang, are then ligated using T4 DNA ligase. This process is gentle and produces high-quality libraries with minimal DNA damage. Ligation kits are recommended for most applications, particularly those requiring high accuracy.

**Rapid kits** (e.g., SQK-RBK114) use a transposase complex to fragment the DNA and attach adapters in a single step. The transposase, a hyperactive Tn5 variant, simultaneously cuts the DNA and ligates the adapters. This is faster and requires less input DNA, but it introduces a bias: the transposase has a sequence preference, and the resulting fragments are not uniformly distributed across the genome. More importantly for accuracy, the transposase reaction can leave nicks or damaged bases at the adapter junctions, which may cause the motor protein to stall or the basecaller to produce errors at the read ends. In practice, rapid kits produce slightly lower accuracy than ligation kits, particularly for the first 50–100 bases of each read.

The choice of kit also affects read length. Ligation kits can produce reads exceeding 100 kb with high-molecular-weight DNA, while rapid kits typically produce shorter fragments (5–20 kb). Read length is not directly an accuracy metric, but longer reads improve consensus accuracy by providing more contiguous coverage and reducing the number of breaks in the assembly graph.

### Flow Cell Types and Pore Versions

The flow cell is the consumable that contains the nanopores. ONT has released several pore versions, each with distinct accuracy characteristics.

**R9.4.1** is the most widely used pore. It has a 4-mer sensing region and produces raw read accuracy of 92–95% with current basecallers. Its main weakness is homopolymer resolution: runs of 5 or more identical nucleotides are frequently miscounted. R9.4.1 is being phased out but remains in use for many legacy datasets.

**R10.4.1** is the current standard. It has a 9-mer sensing region, achieved by fusing two CsgG pores in tandem. The longer sensing region provides more information per nucleotide, dramatically improving homopolymer resolution. Raw read accuracy is 95–98% with the latest basecallers. R10.4.1 also has improved discrimination of modified bases, particularly 5mC and 5hmC.

**R10.4.1 with the "duplex" option** is a special mode where both strands of a double-stranded DNA molecule are sequenced sequentially. The two reads are then combined into a single, high-accuracy consensus. Duplex sequencing achieves raw read accuracy of 99–99.9% (Q20–Q30) per read, approaching short-read accuracy. The trade-off is throughput: only a fraction of pores produce duplex reads, and the sequencing time is doubled.

Flow cell versions also matter. The MinION flow cell (R9.4.1 or R10.4.1) contains 512 channels, while the PromethION flow cell contains 2,675 channels. The PromethION flow cell has a higher density of pores and a more stable temperature, which reduces signal drift. In practice, PromethION data tends to have slightly higher accuracy than MinION data from the same library, likely due to improved thermal and electrical stability.

The sequencing buffer also plays a role. ONT provides a proprietary buffer (e.g., "Sequencing Buffer II") that contains salts, a reducing agent, and a surfactant. The buffer's ionic strength affects the baseline current and the signal-to-noise ratio. Using expired or incorrectly prepared buffer can degrade accuracy. The recommended buffer concentration is typically 50 mM KCl, 5 mM HEPES, and 1 mM EDTA at pH 8.0, though the exact formulation is proprietary.

## Strategies to Improve Accuracy: Consensus and Hybrid Approaches

Even with the best pore and basecaller, raw read accuracy is insufficient for many applications. Several strategies can push accuracy to the level required for clinical or reference-quality genomics.

### Duplex Sequencing

Duplex sequencing is the most direct way to improve accuracy on a single molecule. In this approach, the sequencing adapter is designed to allow both strands of the double-stranded DNA to be sequenced sequentially through the same pore. The motor protein first threads the template strand, then the complement. The two resulting reads are aligned to each other, and a consensus is called. Because errors are largely stochastic, the consensus of two independent reads of the same molecule eliminates most errors.

The accuracy of duplex reads depends on the error rate of the individual strands. With R10.4.1, each strand has ~95–98% accuracy, so the duplex consensus achieves ~99.5–99.9%. This is sufficient for most variant calling applications, including detecting single-nucleotide variants and small indels. Duplex sequencing is particularly valuable for detecting low-frequency variants in heterogeneous samples, such as tumor biopsies or liquid biopsies, where a single mutated molecule must be distinguished from a large background of wild-type DNA.

The main limitation of duplex sequencing is yield. In a typical run, only 5–20% of pores produce duplex reads; the rest produce single-strand (simplex) reads. This is because the motor protein must successfully complete the first strand, then re-thread the complement without dissociating. The duplex yield can be improved by using a "duplex-enhanced" adapter and by optimizing the motor protein concentration, but it remains a significant throughput penalty.

### Hybrid Assembly with Illumina Reads

For de novo assembly and [structural variant](/blog/guides/structural-variant) detection, hybrid approaches that combine nanopore long reads with Illumina short reads offer a powerful accuracy boost. The nanopore reads provide long-range contiguity, spanning repeats and structural variants that are invisible to short reads. The Illumina reads provide high per-base accuracy (Q30+), which can be used to polish the nanopore assembly.

The standard workflow is as follows:

1. **Assemble** the nanopore reads into contigs using a long-read assembler such as Flye, Raven, or Canu. These assemblers are designed to handle the high error rate of nanopore reads and produce a draft assembly.
2. **Polish** the draft assembly with the nanopore reads themselves using a tool like Medaka or Racon. This step corrects errors using the signal-level information from the nanopore reads.
3. **Polish again** with Illumina short reads using a tool like Pilon or NextPolish. This step corrects residual errors, particularly homopolymer errors and small indels, using the high-accuracy short reads.

This hybrid approach routinely produces assemblies with consensus accuracy exceeding Q50 (99.999%) for bacterial genomes and Q40 for human chromosomes. The cost is the need for two sequencing platforms, but for many projects this is justified by the accuracy gain.

It is worth noting that hybrid assembly is not the same as hybrid error correction of individual reads. In hybrid error correction, the short reads are used to correct errors in each nanopore read before assembly. This approach is computationally intensive and has largely been superseded by the assemble-then-polish workflow, which is more efficient and produces better results.

## Common Pitfalls

Even with optimal protocols, several common mistakes can degrade accuracy. Being aware of these failure modes can save time and money.

**Using the wrong basecalling model.** Each pore version and chemistry has a specific basecalling model. Using an R9.4.1 model on R10.4.1 data, or vice versa, produces garbage. Always check the model name against the flow cell and kit used. ONT's Dorado basecaller automatically selects the correct model based on the sequencing summary file, but manual basecalling with Guppy requires explicit model selection.

**Overloading the flow cell.** Loading too much DNA saturates the pores, causing multiple DNA molecules to compete for the same pore. This produces chimeric reads (two different molecules joined together) and increases the frequency of pore blockages. The recommended loading is 50–100 fmol of library for a MinION flow cell. Overloading is a common cause of poor accuracy and low yield.

**Skipping the DNA damage repair step.** The end-repair step in ligation kits includes a damage repair enzyme mix (e.g., NEBNext FFPE DNA Repair Mix) that removes abasic sites, nicks, and oxidized bases. Skipping this step increases the frequency of motor protein stalling and basecalling errors, particularly in FFPE or ancient DNA samples. The damage repair step is essential for high-accuracy results.

**Ignoring read quality filtering.** Not all reads are equal. Reads from damaged pores, reads that are too short (<500 bp), and reads with low Q-scores should be filtered out before downstream analysis. A simple filter for Q-score ≥10 and read length ≥1 kb will remove most low-quality data and improve consensus accuracy. For duplex reads, only use reads where both strands pass quality filters.

**Using an outdated basecaller.** ONT releases new basecalling models frequently, and each update typically improves accuracy. If you are using a basecaller that is more than six months old, you are likely leaving accuracy on the table. Check the ONT community forum for the latest model releases and benchmark results.

**Confusing raw read accuracy with consensus accuracy.** A 95% raw read accuracy sounds low, but with 30× coverage, the consensus accuracy can exceed Q50. Do not judge the platform's suitability for your application based on raw read accuracy alone. Run a small test experiment and measure the consensus accuracy on a known reference before committing to a full-scale run.

## Frequently Asked Questions

### What is the current accuracy of Oxford Nanopore sequencing?

With the R10.4.1 pore and the latest Dorado basecaller, raw read accuracy is typically 95–98% for standard ligation libraries. Duplex sequencing achieves 99–99.9% raw read accuracy. Consensus accuracy depends on coverage: 30× coverage of a bacterial genome typically yields Q50 (99.999%) consensus accuracy, while human-scale genomes with 50× coverage achieve Q40 (99.99%). These numbers improve with each basecaller update, so check the latest ONT benchmarks for current figures.

### How does Oxford Nanopore accuracy compare to Illumina?

[Illumina sequencing](/knowledge/diagnostics/molecular/illumina-sequencing-principle-chemistry-and-workflow) achieves raw read accuracy of 99.9% (Q30) or higher, which is superior to nanopore raw reads. However, Illumina reads are short (150–300 bp), which limits their ability to resolve repeats, structural variants, and phasing. Nanopore consensus accuracy can match or exceed Illumina accuracy for most applications, and the long reads provide information that Illumina cannot. For de novo assembly, nanopore (or hybrid) approaches produce more contiguous and complete assemblies. For targeted variant calling in clinical settings, Illumina remains the gold standard due to its higher raw read accuracy and lower cost per base, but nanopore is increasingly competitive.

### What causes errors in Oxford Nanopore sequencing?

Errors arise from several sources: homopolymer miscounting (particularly in R9 pores), base modifications that confuse the basecaller, motor protein stalling or reversal, sequence context effects in GC-rich regions, and electrical noise from the pore or membrane. Systematic errors—those that occur at the same position in every read—cannot be corrected by coverage alone and require improved chemistry or basecalling models.

### Can Oxford Nanopore sequencing be used for clinical diagnostics?

Yes, but with caveats. The R10.4.1 pore with duplex sequencing achieves accuracy sufficient for detecting single-nucleotide variants and small indels in targeted panels. ONT has received regulatory approval for some diagnostic applications, such as screening for antimicrobial resistance and detecting structural variants in genetic disorders. However, for applications requiring ultra-high accuracy, such as detecting low-frequency somatic mutations in liquid biopsies, the platform still requires careful validation and may need hybrid approaches or very high coverage. The real-time nature of nanopore sequencing is a major advantage in clinical settings where time-to-answer is critical.

### How can I improve Oxford Nanopore sequencing accuracy?

Several strategies are effective: use the R10.4.1 pore and the latest basecalling model; use a ligation kit rather than a rapid kit; perform the DNA damage repair step; sequence at moderate coverage (30–50×) and generate consensus; use duplex sequencing for high-accuracy applications; and polish assemblies with Illumina reads if you have access to a short-read platform. Filtering reads by Q-score and length also improves downstream accuracy.

### What is the difference between raw read accuracy and consensus accuracy?

Raw read accuracy is the accuracy of a single sequencing read, typically 95–98% for current nanopore chemistry. Consensus accuracy is the accuracy of a sequence derived from multiple reads aligned to the same region. Because errors are largely stochastic, consensus accuracy improves with coverage: 10× coverage yields ~Q30, 30× yields ~Q40, and 50× yields ~Q50 for small genomes. Consensus accuracy is the relevant metric for most applications, including assembly and variant calling. Raw read accuracy matters for applications that rely on individual molecules, such as detecting base modifications on a single strand.

## Key Takeaways

- Oxford Nanopore raw read accuracy has improved from ~70% to 95–98% with the R10.4.1 pore and modern basecallers, and duplex sequencing achieves 99–99.9% per read.
- Errors arise from homopolymer miscounting, base modifications, motor protein dynamics, sequence context, and electrical noise; systematic errors require chemistry or algorithmic fixes, not just deeper coverage.
- Basecalling algorithms are the single largest determinant of accuracy; always use the latest model matched to your pore and chemistry.
- Library preparation choices—ligation vs rapid kits, damage repair, and input DNA quality—directly affect error rates and read length.
- Consensus accuracy, not raw read accuracy, is the metric that matters for assembly and variant calling; 30–50× coverage yields Q40–Q50 consensus accuracy.
- Duplex sequencing and hybrid assembly with Illumina reads are the most effective strategies for achieving reference-quality accuracy.
- Common pitfalls include using the wrong basecalling model, overloading the flow cell, skipping damage repair, and conflating raw read accuracy with consensus accuracy.

## Further Reading

- Philip M et al. *Prediction of seafloor ecological state using 16S nanopore sequencing*. Marine pollution bulletin. 2025. [PubMed 40743778](https://doi.org/10.1016/j.marpolbul.2025.118489)
- Wang Y et al. *Nanopore sequencing technology, bioinformatics and applications*. Nature biotechnology. 2021. [PubMed 34750572](https://doi.org/10.1038/s41587-021-01108-x)
- Zhang T et al. *Nanopore sequencing: flourishing in its teenage years*. Journal of genetics and genomics = Yi chuan xue bao. 2024. [PubMed 39293510](https://doi.org/10.1016/j.jgg.2024.09.007)
- Zhang T et al. *The newest Oxford Nanopore R10.4.1 full-length 16S rRNA sequencing enables the accurate resolution of species-level microbial community profiling*. Applied and environmental microbiology. 2023. [PubMed 37800969](https://doi.org/10.1128/aem.00605-23)
- Curry KD et al. *Emu: species-level microbial community profiling of full-length 16S rRNA Oxford Nanopore sequencing data*. Nature methods. 2022. [PubMed 35773532](https://doi.org/10.1038/s41592-022-01520-4)
- Jain M et al. *Nanopore sequencing and assembly of a human genome with ultra-long reads*. Nature biotechnology. 2018. [PubMed 29431738](https://doi.org/10.1038/nbt.4060)

## Related Topics

- [Nanopore vs Sanger Sequencing](/knowledge/molecular-biology/nanopore-vs-sanger-sequencing)
- [Bisulfite Sequencing](/knowledge/molecular-biology/bisulfite-sequencing)
- [ATAC Sequencing](/knowledge/molecular-biology/atac-sequencing)
- [Sequencing Coverage](/knowledge/molecular-biology/sequencing-coverage)
- [CHIP Sequencing](/knowledge/molecular-biology/chip-sequencing)

## Related Clinical & Scientific Guides

* [MAPK Pathway: Mechanism, Function, and Clinical Relevance](/knowledge/molecular-biology/mapk-pathway)
* [Mammalian Cell Culture Bioreactors: A Practical Guide](/knowledge/molecular-biology/mammalian-cell-culture-bioreactor)
* [Nucleotide Formation: Biosynthesis and Assembly of DNA/RNA Building Blocks](/knowledge/molecular-biology/nucleotide-formation)