How to Interpret Methylation Calls from Nanopore Sequencing: Understanding Quality Scores, Confidence Intervals, and Biological Relevance

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

Nanopore sequencing detects DNA methylation directly from raw electrical current signals, producing per-read modification calls that are summarized into methylation frequencies, quality scores, and coverage metrics. Interpreting these outputs correctly requires understanding what each number represents, how caller algorithms generate confidence estimates, and which filtering thresholds are appropriate for your biological question. This article explains the meaning of methylation frequency, quality scores, and coverage, and provides guidelines for downstream analysis, with concrete decision criteria for filtering reliable calls.

Key Takeaways

• Nanopore sequencing detects DNA methylation directly from electrical current signals, producing per-read modification probabilities that are aggregated into methylation frequencies, quality scores, and coverage metrics.
• A methylation frequency of 0.80 means 80 percent of sequenced molecules at that locus were called methylated, and coverage depth directly determines the precision of these frequency estimates.
• The confidence interval around a methylation frequency narrows as coverage increases, with a site of 10 reads and 5 methylated calls yielding a 95 percent confidence interval of approximately 0.19 to 0.81.
• For most applications, a per-read modification probability threshold between 0.5 and 0.8 is reasonable, and consistent threshold application across all samples matters more than the specific value.
• A common minimum coverage threshold is 10 reads per position, though higher thresholds such as 20 or 30 reads provide more stable estimates for clinical or diagnostic applications.

At a Glance

The table below summarizes the core output metrics from nanopore methylation callers and their practical interpretation.

Understanding Nanopore Methylation Detection Principles

Nanopore sequencing identifies base modifications without chemical treatment such as bisulfite conversion. The technology measures changes in ionic current as a DNA strand passes through a protein pore, and modified bases alter the current signature in a sequence-dependent manner. This direct detection approach means methylation information is collected simultaneously with the primary sequence data, enabling combined genomic and epigenomic analysis from a single sequencing run.

The detection principle relies on comparing observed current levels against expected models for unmodified and modified bases. Oxford Nanopore Technologies protocols generate long-range haplotypes, variant calls, and methylation data from the same sequencing output, as demonstrated in clinical studies of dystonia where nanopore data provided methylation information alongside structural variant detection [<a href="#ref-1">1</a>]. The simultaneous collection of sequence and methylation data allows researchers to phase methylation patterns with genetic variants, which is valuable for interpreting allele-specific epigenetic regulation.

The raw signal data are processed by basecallers that produce both nucleotide sequences and per-read modification probabilities. These probabilities reflect how well the observed current signal matches the expected pattern for a methylated versus unmethylated base in a specific sequence context. The most common target in human DNA is 5-methylcytosine in CpG dinucleotides, though the technology can detect other modifications when appropriate models are available.

Quality Scores in Methylation Calls

Per-Read Modification Probabilities

Each read at a given position receives a modification probability from the caller. This probability represents the confidence that the specific molecule was methylated at that site. The probability is derived from the comparison between observed current values and trained models for modified and unmodified bases.

The interpretation of these probabilities requires attention to the sequence context. Deep learning approaches for modification calling typically require training data covering all possible sequence contexts, which can be prohibitively expensive or impossible to obtain for some modifications [<a href="#ref-2">2</a>]. This limitation explains why research focuses predominantly on 5mC in CpG contexts, where training data are more readily available. When interpreting quality scores, you should verify which sequence contexts your caller was trained on and whether your target regions fall within those contexts.

Aggregated Quality Metrics

Most methylation callers aggregate per-read probabilities into a site-level methylation frequency, often accompanied by a quality score. The frequency is calculated as the fraction of reads at that position with a modification probability above a threshold. The quality score typically reflects the separation between the modified and unmodified read populations.

A high quality score at a site with low coverage can still produce unreliable frequency estimates. The confidence interval around a methylation frequency depends on both the number of reads and the consistency of their modification calls. For example, a frequency of 0.50 based on 10 reads has a much wider confidence interval than the same frequency based on 100 reads. You should always examine coverage alongside quality scores when deciding whether to trust a methylation call.

Calibration and Threshold Selection

The choice of per-read probability threshold affects both sensitivity and specificity of methylation calls. A lower threshold captures more potential methylation events but increases false positives. A higher threshold produces more confident calls but may miss genuinely methylated sites where the current signal is ambiguous.

For most applications, a per-read modification probability threshold between 0.5 and 0.8 is reasonable, but the optimal value depends on your biological question and the modification type. If you are comparing methylation levels between conditions, consistent threshold application across all samples is more important than the specific threshold value. Document your threshold choice in your analysis protocol to ensure reproducibility.

Coverage Requirements for Reliable Methylation Calls

Minimum Depth Considerations

Coverage depth directly determines the precision of methylation frequency estimates. With low coverage, stochastic sampling of methylated and unmethylated molecules produces noisy frequency values that may not reflect the true population average. The required coverage depends on the expected methylation level and the effect size you need to detect.

For detecting differentially methylated regions between conditions, higher coverage is needed to distinguish true biological differences from sampling noise. A methylation frequency difference of 10 percent between two conditions requires substantially more coverage to detect reliably than a difference of 50 percent. When planning sequencing experiments, consider the expected methylation variability in your samples and the minimum effect size of interest.

Coverage Variability Across the Genome

Coverage is not uniform across the genome. Repetitive regions, GC-rich areas, and other challenging sequences often have lower coverage than unique regions. This variability means that some genomic positions will have insufficient coverage for reliable methylation calls even when the overall sequencing depth appears adequate.

Tools such as MethylomeMiner allow high-confidence methylation sites to be selected based on coverage and methylation rate thresholds, then assigned to coding or non-coding regions using genome annotation [<a href="#ref-3">3</a>]. This approach helps filter out low-confidence positions before downstream analysis. When interpreting methylation calls, always check coverage at individual positions instead of relying solely on average genome-wide coverage.

Coverage and Confidence Intervals

The confidence interval around a methylation frequency narrows as coverage increases. For a site with 10 reads and 5 methylated calls, the frequency is 0.50 but the 95 percent confidence interval spans approximately 0.19 to 0.81. With 100 reads and 50 methylated calls, the same frequency has a confidence interval of approximately 0.40 to 0.60.

These wide intervals at low coverage explain why many analysis pipelines require minimum coverage thresholds before reporting methylation frequencies. A common minimum is 10 reads per position, though higher thresholds such as 20 or 30 reads provide more stable estimates. The appropriate threshold depends on your downstream analysis and the consequences of false positive or false negative methylation calls.

Methylation Frequency Interpretation

What Methylation Frequency Means Biologically

Methylation frequency at a single position represents the proportion of DNA molecules in your sample that carry a methylation mark at that site. This value can reflect the average methylation state across a cell population, or it can indicate allele-specific methylation if the reads can be phased to parental chromosomes.

In clinical applications, methylation patterns can serve as episignatures that inform on underlying pathogenic mechanisms. Long-read sequencing enables episignature detection that is concordant with microarray-based approaches, and these methylation patterns can be used to assess the pathogenicity of genomic variants [<a href="#ref-4">4</a>]. When interpreting methylation frequencies, consider whether your sample represents a homogeneous cell population or a mixture of cell types with different methylation profiles.

Population-Level Methylation Patterns

Methylation frequencies across multiple positions can be aggregated to identify differentially methylated regions, which are genomic segments where methylation levels differ between conditions. These regions often have biological significance because they may encompass regulatory elements such as promoters or enhancers.

Population-level analysis using pangenome data allows comparison of methylation patterns across multiple bacterial genomes, as supported by tools like MethylomeMiner [<a href="#ref-3">3</a>]. This approach can reveal conserved methylation sites that may have functional roles in gene regulation or genome stability. When interpreting methylation frequencies, consider both individual positions and regional patterns.

Mosaic and Heterogeneous Methylation

Some biological contexts produce heterogeneous methylation patterns where different cells in a population have different methylation states. Nanopore sequencing can reveal apparent mosaic expanded repeats and other complex methylation patterns that are difficult to detect with other methods [<a href="#ref-1">1</a>]. The long reads produced by nanopore sequencing allow methylation calls to be phased to individual molecules, providing information about which methylation patterns co-occur on the same DNA molecule.

This molecule-level resolution is valuable for understanding epigenetic heterogeneity, but it also complicates interpretation. A methylation frequency of 0.50 at a position could indicate that half the cells are fully methylated and half are unmethylated, or that all cells are hemimethylated. Additional information from phased reads or complementary assays may be needed to distinguish these scenarios.

Practical Workflow for Methylation Call Interpretation

Step 1: Assess Sequencing Run Quality

Before interpreting methylation calls, verify that the underlying sequencing data meet quality standards. Check the read length distribution, total output, and basecalling quality. Clinical long-read sequencing studies report average coverage around 34X with median read lengths of 5 kb and N50 values of 16 kb, providing a reference point for expected performance [<a href="#ref-5">5</a>].

Poor sequencing quality produces unreliable methylation calls regardless of downstream filtering. If the run quality is inadequate, consider whether additional sequencing is needed before proceeding with methylation analysis.

Step 2: Verify Caller Configuration

Confirm that your methylation caller was configured appropriately for your modification type and sequence context. Different callers use different models and thresholds, and the output format may vary. Document the caller version, model files, and parameters used for your analysis.

If you are analyzing non-CpG methylation or less common modifications, verify that your caller supports these contexts. The requirement for training data covering all possible sequence contexts means that some modifications cannot be called reliably with existing tools [<a href="#ref-2">2</a>].

Step 3: Apply Coverage Filters

Filter methylation calls based on coverage before interpreting frequencies. The minimum coverage threshold should be established based on your biological question and the expected methylation variability in your samples. For exploratory analyses, a threshold of 10 reads per position provides a starting point. For clinical or diagnostic applications, higher thresholds may be appropriate.

Tools that select high-confidence methylation sites based on coverage and methylation rate can automate this filtering step and integrate with genome annotations to assign filtered sites to functional regions [<a href="#ref-3">3</a>].

Step 4: Examine Quality Score Distributions

Plot the distribution of quality scores across your methylation calls to identify any systematic issues. A bimodal distribution may indicate that the caller is distinguishing between two populations of reads, while a uniform distribution may suggest that quality scores are not informative for your data.

Compare quality scores between known methylated and unmethylated control regions to verify that the caller is performing as expected. If quality scores do not separate known positive and negative controls, the methylation calls may not be reliable.

Step 5: Validate with Independent Methods

When possible, validate a subset of methylation calls using an independent method such as bisulfite sequencing or methylation-specific PCR. This validation is particularly important when methylation calls will be used for clinical decisions or when the biological conclusions depend on specific methylation frequencies.

The concordance between nanopore-based methylation patterns and microarray-based episignatures has been demonstrated in clinical studies, providing confidence that nanopore methylation calls can be used for diagnostic applications when properly validated [<a href="#ref-4">4</a>].

Tools and Workflows for Methylation Analysis

Available Software Options

Several tools support methylation call interpretation from nanopore data. MethylomeMiner provides a straightforward workflow for processing methylation calls, selecting high-confidence sites based on coverage and methylation rate, and assigning sites to genomic features [<a href="#ref-3">3</a>]. The tool supports population-level analysis using pangenome data to compare methylation patterns across multiple genomes.

ReQuant offers an alternative approach that imputes full k-mer based modification models from limited training data [<a href="#ref-2">2</a>]. This algorithm enables modification calling when not all sequence contexts can be obtained, which is particularly useful for less common modifications or organisms with unusual sequence composition.

Reproducible Workflow Considerations

Reproducibility is essential for methylation analysis, particularly when results will be compared across studies or used for clinical decisions. Community standards for reproducible workflows emphasize documentation of software versions, parameters, and data processing steps. The nf-core documentation provides guidance on community pipeline standards, usage, and configuration that can be applied to methylation analysis workflows.

Version control for analysis scripts and careful documentation of parameter choices enable other researchers to reproduce your results. The Carpentries lessons provide foundational training in computing, data management, and version control that supports reproducible analysis practices.

Training Resources for Methylation Analysis

EMBL-EBI's training portal delivers structured, self-paced learning pathways covering its major biological data resources and core bioinformatics analysis methods.

These resources are particularly valuable for researchers transitioning from short-read to long-read sequencing analysis, as the data formats and analysis approaches differ substantially between the technologies.

Common Failure Patterns in Methylation Call Interpretation

Overinterpreting Low-Coverage Calls

A frequent error is treating methylation frequencies from low-coverage positions as reliable quantitative measurements. A frequency of 0.80 from 5 reads is not equivalent to a frequency of 0.80 from 100 reads, yet both may be reported without qualification. This overinterpretation can lead to false biological conclusions, particularly when comparing methylation levels between conditions.

Always report coverage alongside methylation frequencies, and apply minimum coverage thresholds before drawing conclusions. When coverage is insufficient, consider whether additional sequencing or alternative analysis approaches are needed.

Ignoring Sequence Context Effects

Methylation calling accuracy varies by sequence context. The requirement for training data covering all possible sequence contexts means that some contexts will have better model support than others [<a href="#ref-2">2</a>]. Calls in poorly supported contexts may have lower accuracy even when quality scores appear acceptable.

Check whether your caller provides context-specific quality information, and be cautious when interpreting methylation calls in sequence contexts that are underrepresented in training data.

Confusing Per-Read Probability with Site-Level Confidence

Per-read modification probabilities and site-level methylation frequencies represent different levels of confidence. A site with high per-read probabilities may still have an unreliable frequency estimate if coverage is low. Conversely, a site with moderate per-read probabilities may have a reliable frequency if many reads support the call.

Distinguish between these metrics when interpreting results and when reporting confidence in methylation calls.

Failing to Account for Sample Heterogeneity

Methylation frequencies from bulk samples represent averages across potentially heterogeneous cell populations. If your sample contains multiple cell types with different methylation profiles, the frequency at any position reflects the mixture instead of any single cell type. This heterogeneity can obscure biologically meaningful methylation differences.

Consider whether cell sorting, single-cell approaches, or computational deconvolution are needed to address sample heterogeneity in your analysis.

Limitations and Interpretation Boundaries

Technical Limitations of Nanopore Methylation Calling

Nanopore methylation calling has inherent technical limitations that affect interpretation. The accuracy of modification detection depends on the quality of the current signal, the training data available for the modification type, and the sequence context. Some modifications cannot be reliably detected with current tools, and the requirement for comprehensive training data limits the range of modifications that can be studied [<a href="#ref-2">2</a>].

The resolution of methylation calls is also limited by the read length and the ability to assign modifications to specific positions. While nanopore reads are long compared to short-read sequencing, the current signal reflects a window of sequence context, and precise modification positioning can be challenging in some regions.

Biological Interpretation Boundaries

Methylation frequencies provide information about the proportion of methylated molecules at a position but do not directly indicate the functional consequence of that methylation. The relationship between methylation and gene expression is context-dependent, and methylation at a position may or may not affect transcription depending on the genomic location and the regulatory elements involved.

Episignature analysis can provide diagnostic information by identifying disease-specific genome-wide methylation changes, but the interpretation of these patterns requires careful comparison to reference datasets and consideration of the clinical context [<a href="#ref-4">4</a>].

When to Escalate to Professional Expertise

Some methylation analysis situations warrant consultation with specialized expertise. If you are interpreting methylation calls for clinical diagnosis, if your analysis involves rare modifications or unusual organisms, or if your results will guide treatment decisions, consider involving a specialist in epigenetic analysis.

Clinical studies have demonstrated that long-read sequencing with methylation analysis can provide diagnostic value in complex cases, but the interpretation of these results requires appropriate expertise and validation [<a href="#ref-1">1</a>][<a href="#ref-5">5</a>]. The integration of methylation data with variant interpretation, as demonstrated in studies of developmental disorders and dystonia, requires careful consideration of multiple data types [<a href="#ref-4">4</a>].

Records and Documentation for Methylation Analysis

Essential Records for Reproducible Analysis

Maintain detailed records of your methylation analysis to ensure reproducibility and enable troubleshooting. Essential records include the sequencing platform and version, basecaller and methylation caller versions, model files used, parameter settings, and filtering thresholds applied.

Document the coverage and quality score distributions for each sample, as these metrics provide context for interpreting methylation frequencies. Record any deviations from standard protocols and the rationale for those deviations.

Quality Control Metrics to Track

Track quality control metrics throughout your methylation analysis to identify potential issues early. Key metrics include the number of reads passing quality filters, the distribution of read lengths, the coverage across the genome, and the proportion of positions meeting minimum coverage thresholds.

Monitor the concordance between replicate samples and between different analysis approaches to verify the reliability of your methylation calls. Discrepancies between replicates may indicate technical issues or genuine biological variability that requires investigation.

Safety and Ethical Considerations

Data Management and Privacy

Methylation data from human samples contain sensitive biological information that requires appropriate data management and privacy protections. Follow institutional guidelines for data storage, sharing, and publication. When working with clinical samples, ensure that appropriate consent and ethical approvals are in place.

The NCBI provides data resources and search systems that support responsible data sharing and analysis. When depositing methylation data in public databases, ensure that the data are appropriately de-identified and that deposition complies with relevant regulations and institutional policies.

Clinical Interpretation Responsibilities

When methylation data are used for clinical interpretation, the responsible parties must ensure that the analysis meets appropriate standards for diagnostic use. This includes validation of the methylation calls, appropriate quality control, and interpretation by qualified professionals.

The use of methylation episignatures for variant classification and diagnosis requires careful validation and comparison to established reference data [<a href="#ref-4">4</a>]. Clinical decisions should not be based solely on methylation data without appropriate corroborating evidence and professional interpretation.

Building a Methylation Call Decision Framework: From Raw Output to Biological Interpretation

Interpreting nanopore methylation calls requires more than understanding individual metrics. You need a structured decision framework that guides you from raw caller output to defensible biological conclusions. This section provides a practical framework for evaluating methylation calls, establishing confidence thresholds, documenting decisions, and troubleshooting common interpretation failures. The framework is designed to work across different caller outputs and biological questions while maintaining reproducibility and transparency.

Establishing Your Methylation Call Confidence Thresholds

Defining Confidence Tiers for Different Analysis Purposes

Not all methylation calls require the same level of confidence. A screening experiment to identify candidate differentially methylated regions can tolerate more false positives than a clinical validation study or a mechanistic investigation of a specific regulatory element. Establishing confidence tiers before you begin analysis prevents arbitrary threshold changes that can bias results.

Define three confidence tiers based on your analysis goals. The exploratory tier suits hypothesis generation and initial screening, where you accept higher false positive rates to avoid missing potentially interesting regions. The standard tier fits most comparative analyses between conditions, balancing sensitivity and specificity. The stringent tier applies to clinical interpretation, validation studies, or any analysis where false positive methylation calls have significant consequences.

For each tier, specify the minimum coverage, per-read modification probability threshold, and site-level quality score. Document these thresholds in your analysis protocol before examining the data. Pre-registering thresholds prevents the common failure pattern of adjusting filters until results match expectations, which inflates false discovery rates and undermines reproducibility.

Context-Specific Threshold Adjustments

Methylation calling accuracy varies by sequence context, and your confidence thresholds should reflect this variation. The requirement for training data covering all possible sequence contexts means some contexts have stronger model support than others [<a href="#ref-2">2</a>]. When your caller provides context-specific quality information, use it to adjust thresholds instead of applying uniform cutoffs across the genome.

For CpG contexts in human DNA, where training data are most abundant, you can apply standard thresholds with confidence. For non-CpG contexts or less common modifications, increase your coverage requirements and quality score thresholds to compensate for reduced model support. If your caller does not provide context-specific quality metrics, consider whether the sequence composition of your target regions falls within well-supported contexts before proceeding with interpretation.

Threshold Documentation and Justification

Document the rationale for each threshold choice in your analysis records. Include the biological question, the consequences of false positives and false negatives, and any pilot data that informed your decisions. This documentation supports reproducibility and helps other researchers understand why specific thresholds were applied.

When publishing results, report the thresholds used and the number of positions passing each filter. This transparency allows readers to assess the robustness of your findings and compare results across studies. The nf-core documentation provides guidance on community standards for documenting analysis parameters and configurations that can be applied to methylation workflows.

Building a Position-Level Decision Matrix

Categorizing Methylation Calls by Confidence

Create a decision matrix that categorizes each methylation call based on coverage, quality score, and biological context. This matrix provides a structured approach to deciding which calls to trust, which to treat as provisional, and which to exclude from analysis.

The matrix should include four categories. High-confidence calls meet all thresholds for your analysis tier and are suitable for quantitative interpretation. Provisional calls meet coverage requirements but have borderline quality scores or fall in poorly supported sequence contexts, requiring additional validation before drawing conclusions. Low-confidence calls fail one or more thresholds and should be excluded from quantitative analysis, though they may be noted for follow-up. Uninterpretable calls have insufficient coverage or quality to support any methylation assignment and should be treated as missing data instead of as unmethylated positions.

Apply this matrix consistently across all samples and positions. Inconsistent application of confidence categories introduces bias, particularly when comparing methylation levels between conditions where one sample has systematically higher coverage than another.

Handling Borderline Calls

Borderline calls, where coverage meets minimum requirements but quality scores fall near your threshold, require explicit handling rules. Decide in advance whether these calls will be included in downstream analysis, excluded entirely, or flagged for validation. The choice depends on your analysis goals and the tolerance for false positives in your specific application.

For exploratory analyses, including borderline calls with appropriate flagging may be acceptable. For stringent analyses, excluding borderline calls reduces noise and improves the reliability of quantitative comparisons. Document your handling rule and apply it consistently across all samples.

Integrating Biological Context into the Decision Matrix

The decision matrix should incorporate biological context alongside technical metrics. A methylation call in a promoter region with known regulatory function may warrant additional scrutiny or validation compared to a call in an intergenic region with no known function. Conversely, a call in a repetitive region with known alignment challenges may require higher confidence thresholds than a call in a unique region.

Tools such as MethylomeMiner support this integration by assigning high-confidence methylation sites to coding or non-coding regions using genome annotation [<a href="#ref-3">3</a>]. This assignment enables you to prioritize interpretation efforts toward biologically relevant regions while maintaining appropriate skepticism for calls in challenging genomic contexts.

Implementing a Two-Stage Filtering Protocol

Stage One: Technical Quality Filters

The first filtering stage removes positions that cannot support reliable methylation calls due to technical limitations. Apply minimum coverage thresholds, per-read modification probability cutoffs, and quality score minimums as defined by your confidence tiers. This stage produces a set of positions with technically adequate data for methylation interpretation.

Record the number of positions passing and failing each filter. This information helps you assess whether your sequencing depth was adequate for your biological question and whether additional sequencing is needed. If a large proportion of your target regions fail technical filters, the overall sequencing depth may be insufficient for your analysis goals.

Stage Two: Biological Relevance Filters

The second filtering stage selects positions relevant to your biological question. This stage may involve restricting analysis to specific genomic features such as promoters, gene bodies, or differentially methylated region candidates. It may also involve removing positions in known problematic regions such as segmental duplications or regions with alignment ambiguity.

Biological relevance filters reduce the multiple testing burden in genome-wide analyses and focus interpretation efforts on positions most likely to have functional significance. However, applying these filters too aggressively can miss biologically important methylation changes in unexpected locations. Balance the desire for focused analysis with the need for comprehensive discovery.

Iterative Filtering and Sensitivity Analysis

After applying both filtering stages, perform a sensitivity analysis to assess how your results change with different threshold choices. Vary the coverage minimum, quality score cutoff, and per-read probability threshold within reasonable ranges and observe how the number of passing positions and the resulting biological conclusions change.

If conclusions remain stable across a range of thresholds, your results are robust to filtering choices. If conclusions change dramatically with small threshold variations, your data may not support definitive biological interpretation, and additional sequencing or validation may be needed. Document the sensitivity analysis results in your analysis records.

Recording Methylation Call Decisions and Rationale

Structured Decision Logs

Maintain a structured decision log that records every threshold choice, filtering decision, and interpretation judgment made during your analysis. This log should include the date, the analyst, the specific decision, the rationale, and any alternatives considered. Structured decision logs support reproducibility and provide a reference for troubleshooting when results are questioned or when reanalysis is needed.

The decision log complements the technical records described in the existing article by capturing the reasoning behind analysis choices. While technical records document what was done, decision logs document why it was done, which is equally important for reproducibility and for defending analytical choices in peer review or clinical contexts.

Version Control for Analysis Decisions

Use version control for your analysis scripts and decision logs to track changes over time. The Carpentries lessons provide foundational training in version control with Git that supports reproducible analysis practices. Version control enables you to reconstruct exactly which thresholds and decisions produced specific results, which is essential when results are challenged or when you need to compare analyses across different versions of your pipeline.

When using community workflows, document the workflow version and any custom modifications. The nf-core documentation describes standards for workflow versioning and configuration that support reproducibility in genomic analysis.

Linking Decisions to Outputs

Link each decision in your log to the specific outputs it affected. This linkage enables you to trace how threshold choices influenced final results and to assess the impact of each decision on your biological conclusions. When results are questioned, you can quickly identify which decisions contributed to the disputed findings and evaluate whether alternative choices would have produced different conclusions.

This traceability is particularly important in clinical applications where methylation calls may influence variant interpretation or diagnostic decisions. Clinical studies have demonstrated that nanopore-based methylation analysis can direct the interpretation of variants of uncertain significance through episignature profiling [<a href="#ref-1">1</a>][<a href="#ref-4">4</a>]. In these contexts, the ability to trace analytical decisions to specific clinical interpretations is essential for responsible use of the technology.

Troubleshooting Unexpected Methylation Patterns

Systematic Quality Score Anomalies

When quality score distributions deviate from expectations, investigate before interpreting methylation frequencies. A bimodal quality score distribution may indicate the caller is distinguishing between two read populations, which could reflect genuine biological heterogeneity or technical artifacts. A uniform distribution with no clear separation may indicate the caller cannot distinguish modified from unmodified reads in your data.

Compare quality score distributions between known methylated and unmethylated control regions. If quality scores do not separate these controls, the methylation calls may not be reliable regardless of the reported frequencies. This validation step should be performed before extensive downstream analysis to avoid wasting effort on unreliable data.

Coverage-Dependent Frequency Artifacts

Methylation frequencies can show systematic biases at low coverage that are not present at high coverage. If you observe that methylation frequencies correlate with coverage across your dataset, investigate whether this correlation reflects biological reality or technical artifacts. Low-coverage positions may have biased frequencies if the reads that align to those positions are not representative of the underlying methylation state.

This bias can arise from sequence-specific alignment preferences or from PCR amplification biases during library preparation. If coverage-dependent frequency artifacts are present, consider whether additional sequencing or alternative analysis approaches are needed to obtain reliable methylation estimates.

Discordance Between Replicate Samples

Discordance between replicate samples can indicate technical issues or genuine biological variability. When replicates disagree, first verify that the same filtering thresholds and analysis parameters were applied to both samples. Then examine whether the discordant positions share common features such as low coverage, specific sequence contexts, or genomic locations.

If discordance persists after technical factors are ruled out, consider whether the biological samples genuinely differ in methylation patterns. This difference could reflect sample heterogeneity, developmental or environmental influences, or technical differences in sample processing. The resolution of replicate discordance requires careful investigation instead of arbitrary selection of one replicate over another.

Unexpected Methylation in Control Regions

Methylation calls in regions expected to be unmethylated can indicate technical artifacts or genuine biological phenomena. Before concluding that the calls are artifacts, verify that the control regions are truly expected to be unmethylated in your sample type. Some cell types and developmental stages have methylation patterns that differ from common expectations.

If the calls appear to be technical artifacts, investigate potential causes including alignment errors, model mismatches, or contamination. The long-read nature of nanopore sequencing provides advantages in avoiding erroneous calls resulting from confounding pseudogene sequences [<a href="#ref-1">1</a>], but alignment challenges can still occur in repetitive or duplicated regions.

Applying the Decision Framework to Different Biological Questions

Comparative Methylation Analysis Between Conditions

For comparing methylation levels between conditions, the decision framework should prioritize consistent application of thresholds across all samples. Differential methylation analysis requires that coverage and quality filters do not systematically differ between comparison groups, as this difference would bias the comparison.

Before performing differential analysis, verify that the number of positions passing filters is similar across samples and that the coverage distributions are comparable. If one condition has systematically lower coverage, the differential analysis may identify false differences driven by sampling noise instead of biological variation.

Episignature and Diagnostic Applications

Episignature analysis requires particularly stringent quality control because the results may influence clinical interpretation. Clinical studies have demonstrated that nanopore-based episignature detection can be concordant with microarray-based approaches and can direct the interpretation of variants of uncertain significance [<a href="#ref-4">4</a>]. The decision framework for these applications should incorporate validation against known episignature controls and comparison to established reference datasets.

When using methylation calls for variant interpretation, integrate the methylation evidence with other data types instead of relying on methylation alone. The simultaneous collection of genomic and epigenomic data from nanopore sequencing enables this integration, as demonstrated in studies where methylation analysis aided in directing the interpretation of variants [<a href="#ref-1">1</a>][<a href="#ref-5">5</a>].

Bacterial and Microbial Methylation Analysis

Bacterial methylation analysis presents different challenges than human methylation analysis, including the need to interpret methylation in the context of restriction-modification systems and phase variation. The decision framework should incorporate organism-specific considerations, including the expected methylation motifs and the potential for population-level variation in methylation patterns.

Tools such as MethylomeMiner support bacterial methylation analysis by enabling high-confidence site selection and population-level comparison across multiple genomes using pangenome data [<a href="#ref-3">3</a>]. When applying the decision framework to bacterial data, consider whether the methylation calls represent the dominant methylation state in the population or reflect subpopulations with different methylation patterns.

Common Decision Framework Failures and Corrections

Threshold Drift During Analysis

A common failure is adjusting thresholds during analysis when initial results do not match expectations. This threshold drift introduces bias and undermines the validity of downstream conclusions. Prevent threshold drift by pre-registering thresholds in your analysis protocol and documenting any deviations with clear justification.

If you find yourself tempted to adjust thresholds to achieve expected results, step back and investigate whether the unexpected results reflect genuine biology or technical issues. Adjusting thresholds to force expected outcomes is a form of confirmation bias that can lead to false conclusions.

Overly Complex Decision Criteria

While the decision framework provides structure, overly complex criteria can become unmanageable and inconsistently applied. If your decision matrix has too many categories or thresholds, analysts may apply it inconsistently across samples or positions. Simplify the framework to the minimum number of categories and thresholds needed for your analysis goals.

The framework should be simple enough to apply consistently across thousands or millions of positions while capturing the distinctions that matter for your biological question. If the framework becomes too complex, revisit your analysis goals and identify which distinctions are truly necessary.

Ignoring Position-Specific Context

Applying uniform thresholds across all positions without considering sequence context or genomic location can produce misleading results. While uniform thresholds provide consistency, they may be inappropriate for positions in poorly supported sequence contexts or challenging genomic regions.

The decision framework should incorporate position-specific context where available, including sequence context support from the caller and genomic annotation information. This integration enables appropriate confidence assessment for each position while maintaining consistency in the overall decision process.

Failing to Document Decisions

The most common framework failure is failing to document decisions and their rationale. Without documentation, the framework cannot be reproduced, audited, or improved. Make documentation a routine part of the analysis workflow instead of an afterthought.

Use structured formats for decision logs that capture the essential information without requiring excessive effort. The documentation should be sufficient for another researcher to understand why specific thresholds were chosen and how decisions were applied to the data.

Integrating the Decision Framework with Reproducible Workflows

Embedding Decisions in Analysis Pipelines

Embed the decision framework directly into your analysis pipeline instead of applying it as a separate manual step. This embedding ensures consistent application across samples and reduces the risk of human error or inconsistency. Pipeline tools can automate threshold application, decision logging, and sensitivity analysis.

Community workflow standards provide guidance for building reproducible analysis pipelines. The nf-core documentation describes standards for pipeline development, usage, and configuration that support reproducibility. The Galaxy Training Network offers tutorials for building accessible analysis workflows that incorporate quality control and filtering steps.

Automating Decision Logs

Automate decision logging within your pipeline to capture threshold choices, filtering decisions, and quality metrics without requiring manual documentation. Automated logs reduce the burden of documentation and ensure that decisions are recorded consistently across all analyses.

The automated logs should capture the software versions, parameter settings, and filtering outcomes for each analysis run. This information supports reproducibility and enables troubleshooting when results are questioned or when analyses need to be repeated.

Training and Competency Development

Learners can follow curated learning tracks through EMBL-EBI's online training platform, building expertise in database navigation, API usage, and computational analysis.

Invest in training for all team members who will apply the decision framework. Consistent application of the framework requires that all analysts understand the rationale behind thresholds and decisions, beyond the mechanical steps of applying filters.

When to Escalate to Professional Expertise

Complex Clinical Interpretations

If your methylation analysis will influence clinical decisions, consider involving a specialist in epigenetic analysis or clinical genomics. The interpretation of methylation patterns in clinical contexts requires understanding of disease-specific episignatures, variant interpretation frameworks, and the limitations of methylation data for diagnostic purposes.

Clinical studies have demonstrated that nanopore-based methylation analysis can provide diagnostic value in complex cases, including the reclassification of variants of uncertain significance [<a href="#ref-1">1</a>][<a href="#ref-4">4</a>]. However, these interpretations require appropriate expertise and should be conducted within established clinical frameworks.

Unusual or Unexpected Methylation Patterns

If your methylation data show patterns that cannot be explained by your biological knowledge or technical understanding, escalate to someone with deeper expertise in methylation biology or nanopore sequencing technology. Unexpected patterns may indicate novel biological phenomena, technical artifacts, or limitations of the analysis approach that require specialized knowledge to interpret.

Document the unexpected patterns and the context in which they were observed before seeking expert input. This documentation helps the expert understand the situation and provide informed guidance.

Validation and Confirmation Studies

If your methylation calls will be used for publication, clinical decisions, or other high-stakes applications, consider involving experts in the design and interpretation of validation studies. Independent validation using complementary methods strengthens the evidence base for your methylation calls and addresses potential concerns about technical artifacts.

The concordance between nanopore-based methylation patterns and established methods has been demonstrated in clinical studies [<a href="#ref-4">4</a>], but validation requirements vary by application and biological question. Expert input can help design appropriate validation strategies for your specific context.

Frequently Asked Questions

What is the difference between methylation frequency and methylation quality score?

Methylation frequency is the proportion of reads at a position that carry a methylation call, representing the fraction of DNA molecules methylated at that site. The quality score reflects the confidence in individual methylation calls based on how well the observed current signal matches expected patterns for modified versus unmodified bases. A position can have high quality scores but unreliable frequency estimates if coverage is low.

What coverage is needed for reliable methylation frequency estimates?

The required coverage depends on the expected methylation level and the effect size you need to detect. For exploratory analyses, a minimum of 10 reads per position provides a starting point, while higher thresholds such as 20 or 30 reads give more stable estimates. Detecting small differences in methylation between conditions requires higher coverage than detecting large differences.

How should I choose the per-read modification probability threshold?

The optimal threshold depends on your biological question and the modification type. A threshold between 0.5 and 0.8 is reasonable for most applications, with lower thresholds increasing sensitivity and higher thresholds increasing specificity. Consistent threshold application across all samples is more important than the specific threshold value.

Can nanopore methylation calls be used for clinical diagnosis?

Clinical studies have demonstrated that nanopore-based methylation analysis can provide diagnostic value, including episignature detection concordant with microarray-based approaches [<a href="#ref-4">4</a>]. However, clinical use requires appropriate validation, quality control, and interpretation by qualified professionals. Methylation data should be integrated with other evidence instead of used in isolation for clinical decisions.

How do I validate nanopore methylation calls?

Validation can be performed by comparing nanopore methylation calls to independent methods such as bisulfite sequencing or methylation-specific PCR for a subset of positions. Concordance between nanopore-based methylation patterns and established episignature assays has been demonstrated in clinical studies, providing confidence in the technology when properly applied [<a href="#ref-4">4</a>].

What are the limitations of methylation calling in non-CpG contexts?

Methylation calling in non-CpG contexts is limited by the availability of training data covering all possible sequence contexts. Deep learning approaches require comprehensive training data that can be prohibitively expensive or impossible to obtain for some modifications [<a href="#ref-2">2</a>]. Tools like ReQuant address this limitation by imputing modification models from limited training data, but the accuracy in non-CpG contexts may be lower than for CpG methylation.

How should I report methylation frequencies in publications?

Report methylation frequencies with the coverage at each position and the quality score distribution. Specify the caller version, model files, and filtering thresholds used in the analysis. Provide confidence intervals for methylation frequencies where appropriate, particularly for positions with moderate coverage.

Troubleshooting Methylation Call Discrepancies: If nanopore methylation calls deviate from expected patterns, first verify sequencing depth, model version compatibility, and reference genome alignment quality.

If methylation calls do not match expected patterns, first verify the sequencing quality and caller configuration. Check whether the sequence context is supported by the caller's training data and whether coverage is adequate at the positions of interest. If technical issues are ruled out, consider whether the unexpected pattern reflects genuine biological variation, such as sample heterogeneity or novel methylation states.

Related Bioinformatics Guides

• Oxford Nanopore Sequencing: From Sample to Base Calls
• Understanding UMI in Single-Cell Sequencing: What It Is and Why It Matters
• How to Choose a Long-Read Sequencing Platform: PacBio vs Oxford Nanopore
• Genomic Data Analytics: Extracting Biological Insights from Large-Scale Sequencing
• Single-Cell Sequencing Analysis Pipeline: From Raw Data to Biological Insights

References and Further Reading

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