Direct RNA Modification Detection with Nanopore: How to Identify m6A and Other Epitranscriptomic Marks from Raw Signal

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-21

Direct RNA sequencing with Oxford Nanopore Technology reads native RNA molecules without cDNA conversion, preserving chemical modifications that are lost during reverse transcription and PCR amplification. The raw electrical current signal produced as RNA passes through the nanopore contains information about modified bases, because modified nucleotides alter the ionic current profile compared to unmodified counterparts. Researchers can detect N6-methyladenosine (m6A), 5-methylcytosine (m5C), pseudouridine, and inosine directly from this signal using specialized computational tools. This article explains the signal-level features that enable modification detection, describes a practical workflow using tools such as Tombo, m6Anet, Nanocompore, and nanoDoc2, and outlines data preprocessing, quality control, interpretation, and reporting steps for researchers working with direct RNA sequencing data.

Key Takeaways

• Direct RNA modification detection with nanopore sequencing identifies m6A, m5C, pseudouridine, and inosine by analyzing raw ionic current signal changes caused by modified nucleotides.
• The Dorado basecaller includes de novo models for four RNA modifications, achieving 96 to 98 percent accuracy for pseudouridine and 94 to 98 percent for m6A in evaluations.
• Comparative tools like Nanocompore require a matched unmodified control sample and detect modifications by statistically comparing signal distributions between conditions without training data.
• Machine learning tools such as m6Anet and nanoDoc2 classify modified sites from signal features, with nanoDoc2 using a WaveNet-based neural network on 6-mer segmented raw current signals.
• Validate nanopore modification calls with orthogonal methods like LC-MS/MS, and document basecaller version, alignment quality, and replicate design to reduce false positive risk.

At a Glance: Direct RNA Modification Detection Workflow

The choice of analysis approach depends on whether you have a matched unmodified control sample, whether you need single-molecule resolution, and which modification type you target. Comparative methods such as Nanocompore require a control sample and detect modifications by comparing signal distributions between conditions. Machine learning methods such as m6Anet and nanoDoc2 use trained models to classify modified sites directly from signal features. Dorado, the current basecaller from Oxford Nanopore Technologies, includes ionic current models for de novo detection of four RNA modifications: pseudouridine, m6A, m5C, and inosine.

Understanding the Nanopore Signal for RNA Modifications

How Direct RNA Sequencing Preserves Modifications

Direct RNA sequencing differs fundamentally from cDNA-based sequencing approaches. In cDNA sequencing, RNA is reverse transcribed into complementary DNA, and the resulting DNA is amplified and sequenced. This process erases all chemical modifications present on the original RNA molecule. Direct RNA sequencing threads the native RNA strand through a protein nanopore, and the resulting ionic current disruptions are measured in real time. Because the RNA molecule itself passes through the pore, modified nucleotides physically alter the current signature in ways that reflect their chemical structure.

Nanopore direct RNA sequencing has transformed transcriptomics by simultaneously resolving full-length isoforms and detecting base-level modifications - including m⁶A, m⁵C, Ψ, and A-to-I edits - in a single pass.

Signal-Level Features That Report Modifications

The raw signal from nanopore sequencing consists of a continuous measurement of ionic current through the pore. As the RNA molecule translocates through the pore, each combination of nucleotides within the pore creates a characteristic current level. A modified nucleotide, such as m6A, changes the current level compared to the unmodified adenosine at the same position. This current shift is the fundamental signal that modification detection tools exploit.

The basecaller processes this raw signal and assigns nucleotide identities while also producing per-base probability scores. The trace value, a base probability feature added by the Guppy basecalling program during processing of the raw signal, provides additional information useful for modification detection. Tools such as nanoDoc2 use this trace value in a signal segmentation algorithm to identify candidate modified positions.

The magnitude of the current shift depends on the specific modification and its position within the k-mer context. Some modifications produce large current changes that are readily detectable, while others produce subtle shifts that require sophisticated statistical modeling. The position of the modified base within the pore at any given time affects the signal, so the surrounding sequence context influences detectability.

Chemistry Updates and Their Impact on Detection

Oxford Nanopore Technologies updated its direct RNA sequencing technology from RNA002 to RNA004 chemistry. This update included an improved basecaller, Dorado, with increased sequencing accuracy and ionic current models for de novo detection of four RNA modifications. Evaluation of the two chemistries using RNA extracted from a GM12878 B-lymphocyte cell line showed that Dorado results for each modification displayed both global and site-specific differences when compared to RNA002 results.

The Dorado pseudouridine model achieved 96 to 98 percent for both accuracy and F1-score in evaluations using in vitro transcription data and synthetic oligonucleotides. The Dorado N6-methyladenosine model achieved 94 to 98 percent accuracy and 96 to 99 percent F1-score. These performance metrics demonstrate that direct RNA sequencing can simultaneously detect pseudouridine, m6A, m5C, and inosine modifications on individual mRNA strands. However, validation of these calls using orthogonal methods such as liquid chromatography with tandem mass spectrometry is critical for increased confidence.

Core Principles of Modification Detection from Raw Signal

Comparative Detection with Control Samples

Comparative approaches identify modifications by contrasting the signal profile of a sample of interest against a non-modified control sample. Nanocompore, a robust analytical framework developed for this purpose, compares an RNA sample of interest against a non-modified control sample without requiring a training set and allows the use of replicates. This strategy detects different RNA modifications with position accuracy in vitro and has been applied to profile m6A in vivo in yeast and human RNAs, as well as in targeted non-coding RNAs.

The comparative approach works because modified nucleotides alter the distribution of current levels at specific positions along the RNA. By statistically comparing the signal distributions between the modified sample and the unmodified control, Nanocompore identifies positions where the signal differs significantly. This approach does not require prior knowledge of the modification type and can detect multiple types of modifications simultaneously.

The requirement for a matched control sample is a practical consideration. For many experimental systems, an unmodified control can be generated through in vitro transcription of the same RNA sequences. For yeast and human samples, controls may come from strains or cell lines with specific modification machinery knocked out. The availability and quality of the control sample directly affect the sensitivity and specificity of comparative detection.

Machine Learning Classification with Trained Models

Machine learning approaches train models on known modified and unmodified sites, then apply these models to classify candidate positions in new data. m6Anet identifies putative N6-methyladenosine sites using a neural network trained on signal features. nanoDoc2 includes a machine learning algorithm in which a 6-mer segmented raw current signal is analyzed by deep one-class classification using a WaveNet-based neural network.

The training data for these models typically comes from in vitro transcribed RNA with known modification status or from orthogonal validation data such as m6A-seq or LC-MS/MS. The models learn the signal features that distinguish modified from unmodified positions in specific sequence contexts. Once trained, the models can be applied to new datasets to predict modification status at each candidate position.

nanoDoc2 outputs an RNA modification detected by a statistical score at each candidate position. The software includes a novel signal segmentation algorithm based on the trace value, and the core analysis uses deep one-class classification. This approach detects multiple types of RNA modification from nanopore direct RNA sequencing data, making it useful for researchers studying diverse modification types.

De Novo Detection with Basecaller Models

The Dorado basecaller includes ionic current models for de novo detection of four RNA modifications: pseudouridine, N6-methyladenosine, 5-methylcytosine, and inosine. These models operate during basecalling, assigning modification probabilities alongside nucleotide identities. This approach requires no separate analysis step and no control sample, making it the simplest workflow for modification detection.

The de novo models have demonstrated strong performance in evaluations. The pseudouridine model achieved 96 to 98 percent accuracy and F1-score, while the N6-methyladenosine model achieved 94 to 98 percent accuracy and 96 to 99 percent F1-score. These results demonstrate that direct RNA sequencing can simultaneously detect multiple modifications on individual mRNA strands.

However, de novo detection has limitations. The models are trained on specific modification types and may not generalize to all sequence contexts. The accuracy metrics come from controlled evaluations using in vitro transcription data and synthetic oligonucleotides, and performance on complex biological samples may differ. Validation with orthogonal methods remains important for confirming modification calls.

Practical Workflow for Modification Detection

Step 1: Experimental Design and Sample Preparation

The experimental design determines which analysis approaches are feasible. If you plan to use comparative methods such as Nanocompore, you need a matched unmodified control sample. This control should be generated from the same RNA sequences as your experimental sample, ideally through in vitro transcription. For organismal studies, consider whether a knockout strain or cell line lacking the modification machinery is available.

Replicate design is important for statistical power. Nanocompore allows the use of replicates, and the inclusion of biological replicates improves the reliability of modification calls. Plan for at least three biological replicates per condition when feasible, considering the cost and input requirements of direct RNA sequencing.

RNA quality and quantity are critical inputs. Direct RNA sequencing has specific input requirements that vary by protocol and chemistry version. The RNA004 chemistry update included optimized protocols for mRNA, rRNA, tRNA, circRNA, and viral RNA. Assess RNA integrity before library preparation and document the input quantity and quality metrics for each sample.

Step 2: Sequencing and Basecalling

During sequencing, the raw signal is recorded in FAST5 or POD5 format. The choice of basecaller and its version affects downstream analysis. Dorado is the current basecaller from Oxford Nanopore Technologies and includes modification detection models. Guppy, the previous basecaller, adds a trace value during processing of the raw signal that some modification detection tools use.

For modification detection, you need to preserve the raw signal information. Basecalling produces FASTQ files with nucleotide sequences and quality scores, but modification detection requires access to the underlying signal. Ensure that your basecalling output includes the information needed by your chosen analysis tools. Some tools require the raw FAST5 files, while others work with basecalled data that includes per-read signal information.

The basecaller version and model affect modification calling accuracy. The RNA004 chemistry update included an improved basecaller with increased sequencing accuracy. Document the basecaller version, model, and parameters used for each dataset, as these choices affect downstream modification calls.

Step 3: Read Alignment and Preprocessing

After basecalling, align the reads to a reference transcriptome or genome. Splice-aware alignment is necessary for eukaryotic RNA because direct RNA sequencing reads span exon junctions. Minimap2 is a commonly used aligner for long-read data, and GraphMap is another option.

Alignment quality directly affects modification detection. Poorly aligned reads produce unreliable signal-to-position mappings, leading to false modification calls. Filter reads based on alignment quality metrics such as mapping quality, alignment length, and the presence of soft clipping. Document the alignment statistics for each sample, including the number of mapped reads, median read length, and mapping rate.

For modification detection, the alignment must place each read at the correct genomic or transcriptomic position. Chimeric reads, which align to multiple locations, should be handled carefully. Reads with ambiguous alignments should be excluded from modification calling to reduce false positives.

Step 4: Modification Calling

The choice of modification calling tool depends on your experimental design and research question. Table 1 summarizes the key characteristics of major tools.

For comparative analysis with Nanocompore, provide the aligned reads from both the experimental sample and the unmodified control. The tool compares signal distributions at each position and outputs statistical measures of difference. Positions with significant differences are candidate modification sites.

For machine learning approaches, m6Anet applies a pretrained model to identify putative m6A sites. nanoDoc2 requires training and testing of the neural network before prediction. The nanoDoc2 workflow includes signal segmentation, train/test of the neural network, and prediction of RNA modifications present in nanopore direct RNA sequencing data.

For de novo detection with Dorado, modification probabilities are generated during basecalling. Extract the modification calls from the basecalling output and apply appropriate filters based on the probability scores.

Step 5: Filtering and Quality Control

Raw modification calls contain false positives that must be filtered. Apply site-level filters based on read coverage, modification probability, and consistency across replicates. Positions with low coverage provide insufficient evidence for reliable modification calls. The minimum coverage threshold depends on the tool and the modification type, but higher coverage generally improves confidence.

For comparative methods, apply statistical significance thresholds and consider multiple testing correction. The number of positions tested across the transcriptome is large, so correction for multiple comparisons is necessary to control false discovery rate.

For machine learning methods, the output probability scores require threshold selection. The optimal threshold balances sensitivity and specificity and may depend on your research question. If you prioritize recall, use a lower threshold, if you prioritize precision, use a higher threshold.

Cross-replicate consistency provides additional confidence. Modification sites called in multiple independent replicates are more reliable than those called in a single replicate. Document the overlap between replicates and consider requiring modification calls in at least two replicates for downstream analysis.

Step 6: Validation with Orthogonal Methods

Validation of modification calls using orthogonal methods is critical for increased confidence. Liquid chromatography with tandem mass spectrometry (LC-MS/MS) provides quantitative measurement of modified nucleosides in RNA samples. This approach can confirm the presence and abundance of specific modifications at the whole-transcriptome level.

For site-specific validation, consider targeted approaches such as m6A-seq or miCLIP for m6A, bisulfite sequencing for m5C, and CMC-RT for pseudouridine. These methods provide independent evidence for modification at specific positions. The concordance between nanopore-based calls and orthogonal validation data strengthens the confidence in your results.

The evaluation of Dorado modification calling performance used in vitro transcription of genomic DNA and synthetic oligonucleotides to assess accuracy. Similar controlled experiments can be designed to validate modification calls in your specific experimental system. Create RNA standards with known modification status and sequence them alongside your experimental samples to assess the performance of your analysis pipeline.

Data Preprocessing and Quality Assessment

Raw Signal Quality Metrics

Assess the quality of raw signal data before proceeding with modification detection. Key metrics include the number of reads generated, read length distribution, and sequencing throughput. The yield of direct RNA sequencing depends on the chemistry version, library preparation protocol, and input RNA quantity.

The current signal quality affects basecalling accuracy and modification detection. Examine the distribution of current levels and identify any anomalies such as signal drift or pore blockages. The sequencing run report provides information about pore activity, read quality, and throughput over time.

For modification detection, the number of reads covering each transcript position determines the statistical power. Transcripts with low expression produce fewer reads and therefore less reliable modification calls. Document the coverage distribution across the transcriptome and identify regions with insufficient coverage for reliable modification detection.

Basecalling Quality Assessment

Basecalling accuracy directly affects modification detection. The RNA004 chemistry update included an improved basecaller with increased sequencing accuracy compared to RNA002. Assess basecalling quality using metrics such as read accuracy, which compares basecalled sequences to a reference.

The trace value, a base probability feature added by Guppy during processing of the raw signal, provides information about basecalling confidence. Tools such as nanoDoc2 use this trace value for signal segmentation. If you plan to use nanoDoc2, ensure that your basecalling output includes the trace value information.

For modification detection with Dorado, the basecaller produces modification probabilities alongside nucleotide calls. Assess the distribution of modification probabilities to understand the confidence of modification calls. The performance metrics for Dorado modification models, including accuracy and F1-score, provide benchmarks for expected performance.

Alignment Quality Assessment

Alignment quality determines the reliability of signal-to-position mapping. Assess alignment statistics including mapping rate, median alignment identity, and the distribution of alignment lengths. Low mapping rates may indicate problems with the reference sequence, library preparation, or basecalling accuracy.

For splice-aware alignment, examine the distribution of intron lengths and the number of reads spanning exon junctions. Direct RNA sequencing reads are full-length, so they should span complete transcripts including exon junctions. Unexpected alignment patterns may indicate issues with the reference annotation or RNA quality.

Filter reads with poor alignment quality before modification calling. Common filters include minimum mapping quality, minimum alignment identity, and removal of secondary alignments. The specific thresholds depend on your data and research question, but conservative filtering reduces false positive modification calls.

Tools for Modification Detection

Tombo for Signal-Level Analysis

Tombo is a tool for analyzing raw nanopore signal data. It provides functions for signal normalization, alternative base detection, and modification detection. Tombo can identify modified bases by comparing the signal profile of each read to the expected signal based on the reference sequence.

The Tombo workflow involves re-squiggling, which maps the raw signal to the reference sequence, and then comparing observed signal levels to expected levels. Positions where the observed signal deviates from expected are candidate modification sites. Tombo can detect multiple types of modifications without requiring a control sample.

Tombo requires access to the raw signal data in FAST5 format. The tool provides commands for data preprocessing, including signal normalization and event detection. After re-squiggling, Tombo outputs per-read statistics that can be aggregated to identify consistently modified positions.

m6Anet for m6A Detection

m6Anet is a neural network-based tool specifically designed for detecting N6-methyladenosine from nanopore direct RNA sequencing data. The tool uses a multi-instance learning approach to classify each read as modified or unmodified at candidate positions.

The m6Anet workflow requires aligned reads and access to the raw signal. The tool processes the signal data for each read at each candidate position and applies the trained neural network to predict modification status. The output includes per-read modification probabilities and per-site aggregated statistics.

m6Anet has been used to identify putative m6A sites in various organisms. The tool was evaluated in the comparison of RNA002 and RNA004 chemistries, where it was run to identify putative N6-methyladenosine sites. The results showed differences between the two chemistries, highlighting the importance of considering chemistry effects when interpreting m6Anet results.

Nanocompore for Comparative Detection

Nanocompore is a robust analytical framework that identifies modifications by comparing an RNA sample of interest against a non-modified control sample. The tool does not require a training set and allows the use of replicates, making it accessible for researchers without extensive machine learning expertise.

The Nanocompore workflow requires aligned reads from both the experimental sample and the control. The tool compares the signal distributions at each position and outputs statistical measures of difference. Nanocompore can detect different RNA modifications with position accuracy in vitro and has been applied to profile m6A in vivo in yeast and human RNAs.

The comparative approach of Nanocompore provides several advantages. It does not require prior knowledge of the modification type, so it can detect unexpected modifications. The use of replicates improves statistical power and reliability. However, the requirement for a matched control sample may be challenging for some experimental systems.

nanoDoc2 for Multiple Modification Types

nanoDoc2 is a software tool for detecting multiple types of RNA modification from nanopore direct RNA sequencing data. The tool includes a novel signal segmentation algorithm based on the trace value, a base probability feature added by the Guppy basecalling program during processing of the raw signal.

The core of nanoDoc2 includes a machine learning algorithm in which a 6-mer segmented raw current signal is analyzed by deep one-class classification using a WaveNet-based neural network. The output is an RNA modification detected by a statistical score at each candidate position.

The nanoDoc2 workflow includes signal segmentation, training and testing of the neural network, and prediction of RNA modifications present in nanopore direct RNA sequencing data. The tool requires users to train the model on their specific data, which may require substantial computational resources and expertise.

Dorado for De Novo Detection

Dorado is the current basecaller from Oxford Nanopore Technologies and includes ionic current models for de novo detection of four RNA modifications: pseudouridine, N6-methyladenosine, 5-methylcytosine, and inosine. The tool operates during basecalling, producing modification probabilities alongside nucleotide calls.

The Dorado pseudouridine model achieved 96 to 98 percent for both accuracy and F1-score in evaluations. The N6-methyladenosine model achieved 94 to 98 percent accuracy and 96 to 99 percent F1-score. These performance metrics demonstrate the capability of direct RNA sequencing to detect multiple modifications simultaneously.

The de novo approach of Dorado requires no control sample and no separate analysis step. This simplicity makes it attractive for researchers new to modification detection. However, the built-in models are limited to the four modification types they were trained on, and validation with orthogonal methods remains important.

Records and Measurements for Reproducible Analysis

Documenting Experimental Parameters

Reproducible modification detection requires comprehensive documentation of experimental parameters. Record the chemistry version (RNA002 or RNA004), library preparation protocol, input RNA quantity and quality, and sequencing run parameters. These factors affect the raw signal and therefore the modification calls.

Document the basecaller version, model, and parameters used for each dataset. The choice of basecaller affects both nucleotide accuracy and modification detection. Dorado and Guppy produce different outputs, and the trace value added by Guppy is required by some analysis tools.

Record the version and parameters of all analysis tools used in the workflow. Tool versions change over time, and updates may affect modification calls. The comparison of RNA002 and RNA004 chemistries showed that Dorado results for each modification displayed both global and site-specific differences when compared to RNA002 results, highlighting the importance of documenting tool versions.

Maintaining Analysis Logs

Maintain detailed logs of all analysis steps, including commands used, parameters set, and intermediate outputs. This documentation enables reproducibility and troubleshooting. If modification calls change after a tool update, the logs help identify the source of the change.

Record the filtering thresholds applied at each step, including read quality filters, alignment filters, and modification probability thresholds. The choice of thresholds affects the final modification calls, and different thresholds may be appropriate for different research questions.

Document the computational environment, including operating system, software versions, and hardware specifications. Machine learning tools such as m6Anet and nanoDoc2 may produce different results on different hardware due to numerical precision differences.

Storing Raw and Processed Data

Store raw signal data in FAST5 or POD5 format, as these files contain the information needed for modification detection. The raw data should be preserved for reanalysis with updated tools and models. The large size of raw signal files requires substantial storage capacity.

Store processed data including basecalled FASTQ files, aligned BAM files, and modification call tables. These intermediate files enable reanalysis without re-running the full pipeline. Document the file formats and the tools used to generate each file.

Consider depositing data in public repositories such as NCBI databases. NCBI's documentation portal covers the technical architecture, record format standards, and programmatic API access patterns for each of its major biological databases. Public data deposition enables validation by other researchers and contributes to the development of improved analysis methods.

Common Failure Patterns and Troubleshooting

Low Read Coverage at Target Positions

Insufficient read coverage at candidate modification sites is a common problem. Direct RNA sequencing produces fewer reads than cDNA-based approaches, and coverage varies across transcripts based on expression level. Positions with low coverage provide unreliable modification calls.

Troubleshooting strategies include increasing sequencing depth, enriching for target transcripts, and using targeted sequencing approaches. The RNA004 chemistry update included optimized protocols for mRNA, rRNA, tRNA, circRNA, and viral RNA, which may improve coverage for specific RNA classes.

When coverage is insufficient, consider aggregating data across replicates or using statistical methods that account for variable coverage. Document the coverage threshold used for modification calling and report the number of positions excluded due to low coverage.

Basecalling Errors Affecting Modification Calls

Basecalling errors can produce false modification calls. If the basecaller misidentifies a nucleotide, the signal-to-position mapping is incorrect, and the modification detection algorithm may flag the position as modified. The RNA004 chemistry update included an improved basecaller with increased sequencing accuracy, which reduces this source of error.

Assess basecalling accuracy by comparing basecalled sequences to a reference. High error rates suggest problems with library preparation, sequencing conditions, or basecaller configuration. Consider re-basecalling with updated models if accuracy is poor.

The trace value added by Guppy provides information about basecalling confidence. Tools such as nanoDoc2 use this information for signal segmentation. Low trace values may indicate uncertain basecalls that should be treated cautiously in modification detection.

Alignment Artifacts and Chimeric Reads

Alignment artifacts produce incorrect signal-to-position mappings. Chimeric reads, which align to multiple genomic locations, are particularly problematic for modification detection. These reads may be generated during library preparation or may represent genuine biological phenomena such as fusion transcripts.

Filter chimeric reads before modification calling. Tools such as Minimap2 provide flags for secondary and supplementary alignments that can be used for filtering. Document the proportion of reads excluded due to chimeric alignment.

For repetitive regions, reads may align equally well to multiple locations. These multi-mapping reads should be handled carefully, as their signal cannot be unambiguously assigned to a single position. Consider excluding multi-mapping reads from modification calling or using tools that account for mapping uncertainty.

Batch Effects Between Samples

Batch effects can produce spurious modification calls in comparative analyses. Differences in library preparation, sequencing runs, or reagent lots can alter the signal profile independently of biological modification status. These effects are particularly problematic for comparative methods such as Nanocompore.

Include replicates to assess batch effects. If replicates cluster by batch instead of by biological condition, batch effects are present. Consider using statistical methods that model batch effects or including batch as a covariate in the analysis.

The comparison of RNA002 and RNA004 chemistries showed that Dorado results for each modification displayed both global and site-specific differences when compared to RNA002 results. This finding demonstrates that chemistry changes can affect modification calls, and samples sequenced with different chemistries should not be directly compared without careful normalization.

Limitations and Interpretation Boundaries

Detection Sensitivity and Specificity

Modification detection from nanopore signal has inherent limitations in sensitivity and specificity. The magnitude of the current shift produced by a modification depends on the modification type and its sequence context. Some modifications produce subtle shifts that are difficult to distinguish from noise.

The performance metrics for Dorado modification models, including 96 to 98 percent accuracy and F1-score for pseudouridine and 94 to 98 percent accuracy and 96 to 99 percent F1-score for m6A, come from controlled evaluations. Performance on complex biological samples may differ due to the diversity of sequence contexts and the presence of multiple modifications on individual RNA molecules.

The co-occurrence of multiple modified residues on individual RNA molecules presents a challenge for detection. When two modifications are close together, their signal effects may interact, making individual detection difficult. Nanocompore has provided novel insights on the co-occurrence of multiple modified residues on individual RNA molecules, but this remains a challenging analysis scenario.

Stoichiometry and Quantitative Interpretation

Nanopore direct RNA sequencing provides information about the proportion of reads carrying a modification at each position. This stoichiometric information is a unique advantage of the technology. However, the relationship between the fraction of modified reads and the true modification stoichiometry is not always straightforward.

The detection threshold affects stoichiometry estimates. Modifications present at low stoichiometry may be missed, leading to underestimation of modification levels. Conversely, false positive calls inflate apparent stoichiometry. The choice of detection threshold therefore directly affects quantitative interpretation.

Validation with orthogonal methods such as LC-MS/MS provides quantitative measurement of modified nucleosides. Comparing nanopore-based stoichiometry estimates with LC-MS/MS measurements can calibrate the relationship between detection frequency and true modification levels.

Transcriptome Coverage and Representation

Direct RNA sequencing coverage is biased toward highly expressed transcripts. Lowly expressed transcripts may have insufficient coverage for reliable modification detection. This coverage bias limits the comprehensiveness of modification profiling.

The input requirements of direct RNA sequencing may limit its application to samples with sufficient RNA quantity. The RNA004 chemistry update included optimized protocols for various RNA classes, but input requirements remain a consideration for limited samples.

The technology captures full-length transcripts, providing information about modification co-occurrence along individual RNA molecules. However, the read length is limited by RNA integrity, and degraded RNA produces shorter reads that may not span complete transcripts.

Validation Requirements

Validation of modification calls using orthogonal methods is critical for increased confidence. The evaluation of Dorado modification calling performance used in vitro transcription of genomic DNA and synthetic oligonucleotides to assess accuracy. Similar controlled experiments can validate modification calls in specific experimental systems.

LC-MS/MS provides quantitative measurement of modified nucleosides at the whole-transcriptome level. This approach confirms the presence and abundance of specific modifications but does not provide site-specific information. Site-specific validation requires targeted approaches such as m6A-seq or miCLIP for m6A.

The concordance between nanopore-based calls and orthogonal validation data strengthens confidence in results. Discrepancies between methods should be investigated, as they may indicate false positives in the nanopore analysis or limitations of the orthogonal method.

Professional Escalation Criteria

When to Seek Specialized Support

Modification detection from nanopore signal requires specialized expertise in bioinformatics and epitranscriptomics. If your analysis produces unexpected results or you encounter technical challenges, consider seeking support from experts in these fields.

EMBL-EBI's training portal delivers structured learning curricula spanning its major genomics, proteomics, and structural biology databases alongside applied bioinformatics analysis methods. The Galaxy Training Network offers accessible workflow training, analysis tutorials, and reproducibility context. These resources can help researchers develop the skills needed for modification detection analysis.

Software Carpentry and Data Carpentry workshops build core computational literacy - Unix shell, scripted data analysis, and Git version control - that underpins reproducible scientific computing. These skills are essential for reproducible bioinformatics analysis. Researchers without strong computational backgrounds may benefit from this training before undertaking modification detection analysis.

When to Consult Statistical or Computational Experts

Modification detection involves complex statistical modeling and machine learning. If you are unsure about the appropriateness of statistical methods or the interpretation of model outputs, consult a biostatistician or computational biologist.

The choice of multiple testing correction method, the handling of batch effects, and the interpretation of machine learning probabilities all require statistical expertise. Incorrect statistical decisions can lead to false conclusions about modification status.

Bioconductor's vignette-driven documentation model ensures every R package ships with a fully executable analysis workflow demonstrating best-practice reproducible genomics. nf-core enforces reproducibility through containerised pipeline releases, schema-validated parameter documentation, and automated CI testing prior to community publication. These resources support rigorous computational analysis.

When to Validate with Specialized Laboratories

If your modification calls have important biological or clinical implications, consider validating with specialized laboratories. LC-MS/MS analysis requires specialized equipment and expertise. Site-specific validation methods such as m6A-seq require established protocols and experience.

Official NCBI resource guides describe database scope, data provenance, submission requirements, and E-utilities API access for its comprehensive suite of biological databases. These resources support data deposition and retrieval for validation studies.

Before submitting results for publication or clinical use, ensure that modification calls meet the standards of the field. Reviewers and regulators expect validation with orthogonal methods and comprehensive documentation of analysis parameters.

Frequently Asked Questions

What is the difference between direct RNA sequencing and cDNA sequencing for modification detection?

Direct RNA sequencing threads native RNA molecules through a nanopore, preserving chemical modifications that are lost during reverse transcription and PCR amplification in cDNA sequencing. The raw electrical current signal produced as RNA passes through the pore contains information about modified bases, because modified nucleotides alter the ionic current profile compared to unmodified counterparts. cDNA sequencing converts RNA to DNA before sequencing, erasing all chemical modifications present on the original RNA molecule.

Which RNA modifications can be detected with nanopore direct RNA sequencing?

Nanopore direct RNA sequencing can detect multiple modification types including N6-methyladenosine (m6A), 5-methylcytosine (m5C), pseudouridine, and inosine. The Dorado basecaller includes ionic current models for de novo detection of these four modifications. Other tools such as Nanocompore can detect different RNA modifications without requiring a training set, and nanoDoc2 detects multiple types of RNA modification using a machine learning approach.

Do I need a control sample for modification detection?

The requirement for a control sample depends on the analysis tool. Comparative methods such as Nanocompore require a non-modified control sample and compare signal distributions between the experimental sample and the control. Machine learning methods such as m6Anet and nanoDoc2 use trained models and do not require a control sample. Dorado includes built-in models for de novo detection and requires no control sample.

What is the difference between Tombo, m6Anet, Nanocompore, and nanoDoc2?

Tombo analyzes raw nanopore signal data and can detect modified bases by comparing observed signal levels to expected levels based on the reference sequence. m6Anet is a neural network-based tool specifically designed for detecting m6A. Nanocompore is a comparative framework that identifies modifications by comparing an RNA sample against a non-modified control. nanoDoc2 uses a WaveNet-based neural network to detect multiple types of RNA modification from segmented raw current signal.

How does the RNA004 chemistry update affect modification detection?

The RNA004 chemistry update included an improved basecaller, Dorado, with increased sequencing accuracy and ionic current models for de novo detection of four RNA modifications. Evaluation comparing RNA002 and RNA004 chemistries showed that Dorado results for each modification displayed both global and site-specific differences when compared to RNA002 results. The Dorado pseudouridine model achieved 96 to 98 percent accuracy and F1-score, and the m6A model achieved 94 to 98 percent accuracy and 96 to 99 percent F1-score.

How should I validate nanopore-based modification calls?

Validation using orthogonal methods is critical for increased confidence. Liquid chromatography with tandem mass spectrometry (LC-MS/MS) provides quantitative measurement of modified nucleosides in RNA samples. Site-specific validation can use targeted approaches such as m6A-seq or miCLIP for m6A, bisulfite sequencing for m5C, and CMC-RT for pseudouridine. Controlled experiments using in vitro transcription of genomic DNA and synthetic oligonucleotides can assess modification calling performance.

What coverage is needed for reliable modification detection?

The minimum coverage for reliable modification detection depends on the tool, the modification type, and the sequence context. Higher coverage generally improves confidence, and positions with low coverage provide insufficient evidence for reliable modification calls. The statistical power for detecting modifications depends on the number of reads covering each transcript position, and transcripts with low expression produce fewer reads and therefore less reliable modification calls.

Can I detect multiple modifications on the same RNA molecule?

Direct RNA sequencing can simultaneously detect pseudouridine, N6-methyladenosine, 5-methylcytosine, and inosine modifications on individual mRNA strands. The technology captures full-length transcripts, providing information about modification co-occurrence along individual RNA molecules. Nanocompore has provided novel insights on the co-occurrence of multiple modified residues on individual RNA molecules, though detecting multiple modifications close together remains challenging.

Related Bioinformatics Guides

• RNA-Seq Batch Effect Detection and Correction
• RNA-Seq Data Analysis Workflow: From Raw Reads to Insights
• RNA Sequencing Data Analysis: From Raw Reads to Differential Expression
• RNA-Seq vs qPCR: Validation and Comparison
• RNA-Seq vs DNA-Seq: Key Differences and Applications

References and Further Reading

• NCBI Data Resources. National Center for Biotechnology Information.
• EMBL-EBI Training. European Bioinformatics Institute.
• Bioconductor. Bioconductor Project.
• Galaxy Training Network. Galaxy Project.
• nf-core Documentation. nf-core.
• The Carpentries Lessons. The Carpentries.
• RNA modifications detection by comparative Nanopore direct RNA sequencing.. Nature communications, 2021.
• Computational methods for RNA modification detection from nanopore direct RNA sequencing data.. RNA biology, 2021.
• Advances in nanopore direct RNA sequencing and its impact on biological research.. Biotechnology advances, 2025.
• RNA Modification Detection Using Nanopore Direct RNA Sequencing and nanoDoc2.. Methods in molecular biology (Clifton, N.J.), 2023.
• Evaluation of Nanopore direct RNA sequencing updates for modification detection.. bioRxiv : the preprint server for biology, 2025.

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