# Choosing the Right Basecaller for Modified Base Detection in Nanopore Sequencing: A Decision Guide Nanopore sequencing detects DNA modifications directly from the ionic current signal as single molecules pass through a pore, which differs from conventional methods that rely on chemical conversion or amplification artifacts. The basecaller you select determines whether modified bases are reported at all, how accurately they are distinguished from canonical bases, and whether your output files are compatible with downstream methylation analysis tools. This article provides a decision framework for researchers choosing between Guppy, Dorado, and Megalodon, with attention to throughput, accuracy, and compatibility with downstream tools. The practical problem is straightforward: Oxford Nanopore Technologies provides closed-source basecallers that integrate modification detection, while open-source alternatives offer transparency and customization but require more setup. Your choice affects the base calls themselves and the signal-level information preserved in your output files, which determines whether you can reanalyze data with different tools later. This guide walks through the technical distinctions, workflow implications, and decision criteria relevant to laboratory professionals and bioinformatics practitioners. ## At a Glance: Basecaller Comparison for Modified Base Detection | Feature | Guppy | Dorado | Megalodon | |---------|-------|--------|-----------| | Source availability | Closed-source from Oxford Nanopore Technologies | Closed-source from Oxford Nanopore Technologies | Open-source community tool | | Modification calling | Built-in for 5mCpG and 5hmC on supported flowcells | Built-in for 5mCpG, 5hmC, and additional motifs | Per-read modification probabilities via signal-level analysis | | Hardware requirements | GPU recommended for high throughput | GPU required for efficient operation | CPU-capable but slow, GPU support varies | | Output compatibility | FASTQ plus modified base tags in BAM | FASTQ plus modified base tags in BAM | BAM with per-read modification probabilities | | Downstream tool support | Broad support in methylation analysis pipelines | Growing support, including DeepMod2 comparisons | Requires custom parsing for many tools | | Reanalysis flexibility | Limited to stored basecall information | Limited to stored basecall information | Retains signal-level data for reanalysis | | Best use case | Established labs with existing pipelines | New projects requiring speed and modern models | Research groups needing transparency and customization | The table above summarizes the primary distinctions. Guppy has been the standard for several years and remains widely supported in existing workflows. Dorado represents the current generation from Oxford Nanopore Technologies with improved speed and model architecture. Megalodon offers an open-source alternative that processes raw signal data directly, providing modification probabilities that some downstream tools require. ## Understanding Modified Base Detection in Nanopore Sequencing Nanopore sequencing detects DNA methylations from the ionic current signal of single molecules, offering a unique advantage over conventional methods that require bisulfite conversion or antibody enrichment. The electrical current changes as a DNA molecule passes through the pore, and these changes reflect the chemical structure of each base. Modified bases such as 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC) produce characteristic current signatures that differ from unmodified cytosine. The basecaller interprets these current signals and assigns base identities. For modification detection, the basecaller must also identify deviations from expected current patterns that indicate chemical modifications. This dual task requires models trained on datasets containing both modified and unmodified sequences. The training data and model architecture directly influence detection accuracy. A signal processing and deep learning framework for methylation detection using Oxford Nanopore sequencing demonstrated that modified base detection can be performed from ionic current signal data across different flowcell types. The study presented DeepMod2, an open-source framework implementing both bidirectional long short-term memory (BiLSTM) and Transformer models, capable of analyzing POD5 and FAST5 signal files generated on R9 and R10 flowcells. DeepMod2 showed comparable performance to Guppy and Dorado, which remain closed-source methods from Oxford Nanopore Technologies. This finding matters for your basecaller choice because it confirms that open-source alternatives can achieve similar accuracy to proprietary tools. The same study reported a high correlation between reduced representation and whole-genome Nanopore sequencing for methylation detection. This correlation supports the use of adaptive sampling, a software-controlled enrichment method for targeted sequencing, which allows reduced representation methylation sequencing applied to CpG islands or imprinted regions. If your experimental design uses adaptive sampling, your basecaller must handle the resulting data structure appropriately. ## Core Principles of Basecalling for Modification Detection Basecalling converts raw electrical current measurements into nucleotide sequences. For modification detection, the process must preserve information about current deviations that indicate chemical changes to bases. Three principles govern this process. First, the basecaller must operate on raw signal data, not on previously basecalled sequences. Signal-level analysis captures the subtle current differences between modified and unmodified bases. If your pipeline starts from FASTQ files, modification information has already been lost unless the basecaller embedded it in the output format. Second, the model architecture determines the basecaller's ability to learn complex current patterns. Recurrent neural networks, particularly those using connectionist temporal classification, have become the leading approach for nanopore basecalling. A hybrid basecaller named Lokatt demonstrated that incorporating prior biological and physical insights through an explicit duration hidden Markov model combined with a residual LSTM network achieves performance on par with existing state-of-the-art structures. Lokatt achieved a median single read identity score of 0.930 and a genome coverage ratio of 99.750% when trained on methylation-free E. coli samples with MinION R9.4.1 chemistry. This result indicates that model architecture innovations continue to improve basecalling accuracy. Third, the output format must support modification annotations. The BAM format with MM and ML tags has become the standard for storing modified base information. These tags record the position and probability of modified bases within each read. Tools that do not produce these tags require separate processing steps to extract modification information. ## Workflow Options for Modified Base Detection Your basecalling workflow depends on whether you prioritize speed, accuracy, transparency, or compatibility with specific downstream tools. The following sections describe the practical considerations for each basecaller. ### Guppy: The Established Standard Guppy has served as the default basecaller for Oxford Nanopore sequencing platforms for several years. It provides built-in modification detection for common motifs including 5mCpG and 5hmC on supported flowcell and chemistry combinations. The software runs on both CPU and GPU configurations, though GPU acceleration substantially improves throughput for high-volume projects. The closed-source nature of Guppy means you cannot inspect or modify the underlying models. This limitation affects reproducibility if you need to document exactly how base calls were generated. However, Guppy's widespread adoption means most downstream analysis tools accept its output format without modification. For modification detection, Guppy produces BAM files with MM and ML tags when configured appropriately. The basecaller applies a trained model that distinguishes modified from unmodified bases based on current signal patterns. The accuracy of these calls depends on the model version and the match between your sequencing conditions and the conditions used to train the model. Practical considerations for Guppy include the need to select the correct model for your flowcell type and chemistry. Using a mismatched model produces poor basecalling accuracy and unreliable modification calls. Oxford Nanopore Technologies provides model recommendations through their documentation and software interfaces. ### Dorado: The Current Generation Dorado represents the current generation of basecallers from Oxford Nanopore Technologies. It incorporates newer model architectures and provides faster processing compared to Guppy. Dorado supports modification detection for 5mCpG and 5hmC, with additional motifs available depending on the model version. The performance comparison in the DeepMod2 study placed Dorado among the current state-of-the-art methods for methylation detection. The study noted that Dorado and Guppy remain closed-source, which limits independent verification of their underlying algorithms. For practical purposes, Dorado offers improved speed and updated models that may provide better accuracy on newer flowcell types. Dorado requires GPU acceleration for efficient operation. Laboratories without access to suitable GPUs may find Dorado impractical for large datasets. The software produces BAM files with modification tags compatible with downstream methylation analysis tools. One advantage of Dorado is its integration with newer Oxford Nanopore sequencing platforms. If you are using R10 flowcells or the latest chemistry, Dorado likely provides the most current models for your specific conditions. Guppy models may lag behind for newer platforms. ### Megalodon: The Open-Source Alternative Megalodon takes a different approach by performing signal-level analysis to estimate modification probabilities for each read. Instead of producing a single consensus call, Megalodon outputs per-read modification probabilities that can be aggregated for downstream analysis. This approach provides more detailed information but requires additional processing steps. The open-source nature of Megalodon allows researchers to inspect the code, modify parameters, and adapt the tool to specific experimental needs. This transparency supports reproducibility and customization that closed-source tools cannot provide. However, Megalodon requires more computational resources and expertise to operate effectively. Megalodon processes raw signal data directly, which means it can be applied to data from various flowcell types and chemistries. The tool supports both FAST5 and POD5 signal file formats. Output includes BAM files with modification probabilities that can be parsed for downstream analysis. The main limitation of Megalodon is speed. Signal-level analysis is computationally intensive, and processing large datasets can take substantially longer than with Guppy or Dorado. Laboratories with high-throughput requirements may find Megalodon impractical for routine use. ## Decision Tree for Basecaller Selection The following decision framework helps match your experimental requirements to the appropriate basecaller. ### Step 1: Assess Your Throughput Requirements Estimate the total number of reads or bases you need to process. For projects exceeding several million reads, GPU-accelerated basecalling becomes necessary. Dorado provides the fastest processing among the three options. Guppy with GPU acceleration offers comparable throughput for established workflows. Megalodon is suitable only for smaller datasets or targeted analyses where speed is not the primary constraint. ### Step 2: Determine Your Modification Detection Needs Identify which modified bases you need to detect. Guppy and Dorado provide built-in support for common modifications including 5mCpG and 5hmC. If you need to detect less common modifications or custom motifs, Megalodon's open-source framework allows you to train or adapt models for your specific targets. The DeepMod2 study demonstrated that open-source frameworks can achieve comparable performance to Guppy and Dorado for methylation detection, supporting the viability of open-source approaches for specialized needs. ### Step 3: Evaluate Downstream Tool Compatibility Review the analysis tools you plan to use after basecalling. Most established methylation analysis tools accept BAM files with MM and ML tags, which Guppy and Dorado produce. Some tools require per-read modification probabilities that Megalodon provides. Check the documentation for your downstream tools to confirm which basecaller output formats they support. ### Step 4: Consider Reanalysis Flexibility Determine whether you may need to reanalyze your data with different parameters or models in the future. If you retain raw signal files, you can re-run any basecaller at any time. However, if you only store basecalled output, your analysis options are limited to what the original basecaller produced. Megalodon's signal-level analysis preserves more information in its output, but the raw signal files remain the most flexible storage format. ### Step 5: Match Basecaller to Flowcell and Chemistry Verify that your chosen basecaller supports your specific flowcell type and chemistry. Guppy and Dorado provide models for supported combinations, and using the correct model is essential for accurate modification detection. Megalodon can process signal data from various flowcell types, but you may need to configure parameters for your specific conditions. ## Practical Implementation Steps Implementing a basecalling workflow for modified base detection requires attention to data management, software configuration, and quality control. ### Data Input Preparation Raw signal data from Oxford Nanopore sequencing is stored in FAST5 or POD5 formats. Confirm that your basecaller supports the format produced by your sequencing platform. Newer platforms may produce POD5 files, while older systems generate FAST5 files. The DeepMod2 framework supports both formats, indicating that tools increasingly accommodate both file types. Organize your raw data with clear naming conventions that include flowcell identifiers, sample names, and run dates. This organization supports traceability when you need to correlate basecalling results with experimental conditions. ### Software Installation and Configuration Install the basecaller software on a machine that meets the hardware requirements. Guppy and Dorado require specific versions of CUDA for GPU acceleration. Megalodon has different installation requirements depending on whether you use a pre-built package or build from source. Configure the basecaller with the correct model for your flowcell and chemistry. This configuration is critical for modification detection accuracy. Using an incorrect model produces unreliable results that may not be immediately obvious from quality metrics. ### Running the Basecaller Execute the basecaller with parameters appropriate for your data volume and modification detection needs. For Guppy and Dorado, specify the output format to include modification tags. For Megalodon, configure the modification detection parameters to match your target modifications. Monitor the basecalling progress and check for errors or warnings in the log files. Common issues include insufficient disk space, GPU memory errors, and model mismatches. ### Output Validation After basecalling, validate the output files before proceeding with downstream analysis. Check that BAM files contain the expected modification tags and that read counts match the input data. Review quality metrics such as read length distributions and basecall accuracy scores. For modification detection, verify that the proportion of modified bases falls within expected ranges for your sample type. Unusual modification levels may indicate basecaller misconfiguration or sample issues. ## Records and Measurements for Quality Control Maintaining detailed records of your basecalling runs supports reproducibility and troubleshooting. The following measurements should be recorded for each run. ### Basecalling Metrics Record the basecaller version, model version, and configuration parameters for each run. These details allow you to reproduce results and compare across runs. Document the flowcell type, chemistry version, and sequencing platform used for each dataset. Track processing time and resource usage, including CPU and GPU utilization. These metrics help estimate capacity for future projects and identify performance issues. ### Modification Detection Metrics Measure the proportion of reads with modification calls and the distribution of modification probabilities. These metrics indicate whether the basecaller detected modifications consistently across your dataset. Compare modification levels between technical replicates to assess reproducibility. The DeepMod2 study reported high correlation between reduced representation and whole-genome Nanopore sequencing, suggesting that modification detection can be consistent across different experimental approaches when the basecaller performs reliably. ### Quality Metrics Record basecall accuracy metrics such as read identity scores and coverage ratios. The Lokatt study reported a median single read identity score of 0.930 and a genome coverage ratio of 99.750% for their hybrid basecaller, providing benchmarks for evaluating basecalling performance. While these specific values apply to Lokatt, similar metrics for your chosen basecaller help identify quality issues. Monitor the number of reads that fail basecalling or produce low-quality calls. High failure rates may indicate problems with the sequencing run, library preparation, or basecaller configuration. ## Common Failure Patterns and Troubleshooting Several recurring issues affect modified base detection in nanopore sequencing. Recognizing these patterns helps you diagnose problems quickly. ### Model Mismatch Using a basecaller model that does not match your flowcell or chemistry produces poor basecalling accuracy and unreliable modification calls. Symptoms include low read identity scores, unusual modification proportions, and poor genome coverage. Verify that the model version matches your sequencing conditions. ### Insufficient Signal Quality Poor signal quality from degraded DNA, suboptimal library preparation, or flowcell issues reduces basecalling accuracy. Modification detection suffers more than canonical basecalling because modified base signals are subtle. Check sequencing metrics such as pore occupancy and read length distributions to identify signal quality issues. ### Output Format Incompatibility Downstream tools may fail to read modification information from your basecaller output. This failure often occurs when tools expect specific tag formats or when the basecaller version produces tags that differ from what the tool supports. Review the documentation for both the basecaller and downstream tools to confirm compatibility. ### Computational Resource Exhaustion Basecalling large datasets can exhaust memory, disk space, or GPU resources. Symptoms include crashes, slow processing, and incomplete output files. Monitor resource usage during basecalling and allocate sufficient capacity for your dataset size. ### Modification Calling Absent If your output files lack modification tags, the basecaller may not have been configured for modification detection. Check the basecaller parameters to confirm that modification calling is enabled and that the correct modification motifs are specified. ## Limitations of Current Basecallers Understanding the limitations of each basecaller helps you interpret results appropriately and avoid overconfidence in modification calls. ### Closed-Source Constraints Guppy and Dorado are closed-source, meaning you cannot inspect the underlying models or algorithms. This limitation affects reproducibility because you cannot document exactly how base calls were generated. The DeepMod2 study noted this constraint, highlighting that open-source alternatives provide transparency that proprietary tools lack. ### Model Training Dependence All basecallers depend on training data that reflects specific sequencing conditions. If your experimental conditions differ from the training conditions, accuracy may decrease. This dependence is particularly relevant for modification detection, where current signatures vary with sequence context and chemical environment. ### Computational Requirements GPU acceleration is essential for high-throughput basecalling with Guppy and Dorado. Laboratories without GPU resources face substantially longer processing times. Megalodon's signal-level analysis requires even more computational resources, limiting its use for large datasets. ### Modification Detection Scope Current basecallers detect specific modifications for which they have trained models. Less common modifications may not be detected unless you use specialized tools or train custom models. The DeepMod2 framework demonstrates that open-source approaches can extend modification detection to additional contexts, but this capability requires bioinformatics expertise. ## Epigenetic Context and Biological Interpretation Understanding the biological significance of detected modifications helps you interpret basecaller output meaningfully. Epigenetic modifications play critical roles in gene regulation, development, and disease processes. Research on pathogenic fungi demonstrated that epimutations can drive antimicrobial resistance through distinct pathways. In Mucor circinelloides species complex, most FK506-resistant isolates were unstable and transient, reverting to drug sensitivity when passaged without drug, with no associated DNA mutations. Half of the isolates showed RNAi-dependent epimutation where small interfering RNAs silenced the fkbA gene post-transcriptionally. Most of the remaining resistant isolates underwent heterochromatin-mediated silencing via H3K9 dimethylation, transcriptionally repressing fkbA and neighboring genes. A similar mechanism operated in Mucor atramentarius, where FK506 resistance was mediated by ectopic heterochromatin silencing with siRNA spreading across the region. Heterochromatin-mediated fkbA epimutants exhibited stability during in vivo infection, suggesting epimutation could impact pathogenesis. This research context matters for basecaller selection because it demonstrates that modifications beyond 5mCpG and 5hmC can have profound biological consequences. If your research investigates histone modifications, RNAi-mediated silencing, or other epigenetic mechanisms, you need basecalling approaches that accurately capture the relevant DNA modifications. The choice between closed-source and open-source basecallers affects your ability to detect and interpret these biologically significant modifications. ## Training and Skill Development for Basecalling Workflows Developing proficiency in nanopore basecalling requires foundational skills in bioinformatics and computational biology. Several training resources support skill development in these areas. ### Foundational Bioinformatics Training The European Bioinformatics Institute provides training pathways for bioinformatics data resources and practical analysis education. These resources help researchers understand the data structures and analysis approaches relevant to nanopore sequencing. The training covers topics such as sequence databases, file formats, and analysis workflows that apply to basecalling and downstream analysis. ### Reproducible Workflow Training The Galaxy Training Network offers accessible workflow training and analysis tutorials that emphasize reproducibility. These resources help researchers structure their basecalling workflows to produce consistent, documented results. The tutorials cover practical aspects of running analysis tools and managing data across workflow steps. ### Community Pipeline Standards The nf-core documentation describes community pipeline standards for reproducible workflow configuration. These standards apply to nanopore analysis pipelines that incorporate basecalling steps. Following community standards supports collaboration and ensures that your workflows can be shared and reproduced by others. ### Computing Fundamentals The Carpentries lessons provide foundational training in computing, data management, shell, Git, and programming. These skills are essential for operating basecaller software, managing sequencing data, and developing custom analysis scripts. Researchers new to computational work should complete this foundational training before attempting complex basecalling workflows. ### Package and Workflow Documentation The Bioconductor project provides official documentation for packages, workflows, installation, and reproducible genomic analysis. Many downstream methylation analysis tools are available through Bioconductor, and understanding this ecosystem helps you select compatible tools for your basecalling output. ### Data Resource Navigation The National Center for Biotechnology Information provides official descriptions of databases, search systems, sequence resources, and analysis services. These resources help researchers understand the data standards and repositories relevant to nanopore sequencing data deposition and retrieval. ## Integrating Basecalling with Downstream Analysis The basecaller you choose affects the immediate base calls and the compatibility of your data with downstream analysis tools. ### Methylation Analysis Pipelines Most methylation analysis pipelines accept BAM files with MM and ML tags. Guppy and Dorado produce these tags when configured for modification detection. Megalodon produces per-read modification probabilities that require custom parsing for many pipelines. The DeepMod2 study demonstrated that open-source frameworks can achieve comparable performance to Guppy and Dorado for methylation detection. This finding suggests that researchers have viable options beyond the proprietary tools, particularly when they need to analyze data from diverse flowcell types or implement custom modification detection. ### Structural Variant Analysis Long read sequencing data from nanopore platforms supports structural variant detection. The basecaller's accuracy affects structural variant calling because mis-called bases can create false variant signals. Higher accuracy basecallers reduce false positives in structural variant analysis. ### Integration with Other Sequencing Data Nanopore data is often integrated with data from other sequencing platforms, such as PacBio or short-read sequencing. The basecaller choice affects how well nanopore data integrates with these other data types. Consistent basecalling across samples and runs supports reliable integration. ## Cost and Resource Considerations The basecaller choice has practical implications for computational costs and resource allocation. ### Hardware Investment GPU-accelerated basecalling requires investment in suitable hardware. Dorado and Guppy with GPU acceleration provide the fastest processing but require compatible GPUs. Megalodon can run on CPU but requires substantially more processing time. ### Software Licensing Guppy and Dorado are available through Oxford Nanopore Technologies, with usage terms that may vary by institution and application. Megalodon is open-source and freely available. Review the licensing terms for your chosen basecaller to ensure compliance with institutional policies. ### Personnel Time Open-source tools like Megalodon require more expertise and time to configure and operate. Factor personnel time into your cost assessment when choosing between proprietary and open-source basecallers. ## Case-Based Decision Scenarios The following scenarios illustrate how the decision framework applies to common research situations. ### Scenario 1: High-Throughput Whole Genome Methylation Study A research group plans to sequence 100 human genomes for methylation analysis. The project requires consistent modification detection across all samples and compatibility with established methylation analysis pipelines. Dorado provides the fastest processing and current models for the sequencing platform. The group should verify that their downstream tools accept Dorado output and allocate sufficient GPU resources for the project scale. ### Scenario 2: Targeted Methylation Analysis with Adaptive Sampling A laboratory studies methylation patterns in specific genomic regions using adaptive sampling to enrich target loci. The DeepMod2 study demonstrated high correlation between reduced representation and whole-genome nanopore sequencing, supporting this approach. The group can use Guppy or Dorado for basecalling, but should confirm that the basecaller handles the data structure produced by adaptive sampling correctly. ### Scenario 3: Custom Modification Detection in Non-Model Organisms A research group studies epigenetic modifications in a non-model organism with unusual modification patterns. The group needs to detect modifications beyond the standard 5mCpG and 5hmC motifs. Megalodon's open-source framework allows customization for specific modification types. The group should allocate additional time for model development and validation. ### Scenario 4: Reproducibility-Critical Clinical Research A clinical research group needs to document every step of their analysis for regulatory compliance. The closed-source nature of Guppy and Dorado limits documentation of the underlying algorithms. The group may prefer Megalodon for its transparency, accepting the computational cost in exchange for complete documentation capability. ## Quality Control Metrics for Modification Detection Establishing quality control metrics specific to modification detection helps identify problems early in the analysis pipeline. ### Modification Proportion Checks Compare the proportion of modified bases in your samples to expected values from literature or preliminary experiments. Substantial deviations warrant investigation. The DeepMod2 study's correlation between reduced representation and whole-genome sequencing provides a benchmark for consistency across experimental approaches. ### Read-Level Modification Consistency Examine the distribution of modification probabilities across reads. Consistent probabilities suggest reliable detection, while highly variable probabilities may indicate technical issues. Check whether modification calls cluster by read position or quality score. ### Replicate Reproducibility Run technical replicates through your basecalling workflow and compare modification calls. High concordance between replicates supports the reliability of your basecalling approach. Discordant calls may indicate basecaller instability or sample heterogeneity. ### Comparison with Independent Methods If possible, validate modification calls using an independent method such as bisulfite sequencing or antibody-based enrichment. Agreement between methods strengthens confidence in your basecalling results. Disagreement may indicate limitations in either approach. ## Data Storage and Archival Considerations Nanopore sequencing data requires substantial storage capacity, and your basecalling choices affect storage requirements. ### Raw Signal Data Raw signal files in FAST5 or POD5 format are the most flexible storage format because they allow reanalysis with any basecaller. However, these files are large and require significant storage capacity. Develop a storage strategy that balances flexibility with practical constraints. ### Basecalled Output Basecalled BAM files with modification tags are smaller than raw signal files but contain less information. If you store only basecalled output, you cannot reanalyze with different basecallers or parameters. Consider whether your project requires the flexibility of raw signal storage. ### Archival Policies Institutional and funding agency policies may specify data retention requirements. Review these policies when developing your storage strategy. The National Center for Biotechnology Information provides resources for sequence data deposition that may apply to your project. ## Professional Escalation Criteria Certain situations warrant escalation to senior researchers, bioinformatics specialists, or institutional support services. ### Persistent Quality Issues If basecalling quality metrics remain poor despite correct configuration and adequate signal quality, escalate to a bioinformatics specialist. Persistent issues may indicate problems with the basecaller software, model files, or computational environment that require expert diagnosis. ### Unexpected Modification Patterns If modification levels deviate substantially from expected ranges for your sample type, consult with a domain expert before proceeding with downstream analysis. Unexpected patterns may indicate biological phenomena of interest or technical artifacts that require investigation. The research on Mucor fungi demonstrated that transient epimutations can produce biologically meaningful modification changes without DNA sequence mutations, so unusual patterns deserve careful evaluation. ### Software or Pipeline Failures When basecaller software crashes repeatedly or produces corrupted output, escalate to the software provider or community support channels. Document the error messages and your configuration details to facilitate troubleshooting. ### Reproducibility Concerns If you cannot reproduce basecalling results across runs or laboratories, escalate to institutional bioinformatics support. Reproducibility issues may stem from software version differences, model updates, or configuration variations that require systematic investigation. ## Building a Basecaller Validation Protocol for Modified Base Detection Selecting a basecaller is only the first step. The more pressing operational problem is confirming that your chosen basecaller actually detects modifications correctly on your specific samples, flowcells, and chemistry. A structured validation protocol distinguishes genuine biological modification signals from technical artifacts and provides documented evidence that your basecalling pipeline performs as expected. This section presents a practical validation framework that you can implement before committing to large-scale production runs. ### Establishing a Reference Sample Set Validation requires samples with known modification status. Create a reference set that includes three categories: a positive control with known high-level methylation, a negative control with minimal or absent methylation, and a spike-in mixture with defined modification ratios. For CpG methylation studies, commercially available fully methylated and unmethylated control DNA provides a reliable baseline. The DeepMod2 study used multiple publicly available and newly generated datasets to evaluate performance under varying scenarios, demonstrating that diverse reference materials strengthen validation conclusions. For each reference sample, document the expected modification proportion based on the supplier specifications or independent measurement. Store these reference samples under consistent conditions and track their passage history. The Mucor research demonstrated that epigenetic states can be transient, with resistant isolates reverting to sensitivity when passaged without drug. This finding underscores the importance of verifying reference sample modification status periodically instead of assuming stability over time. ### Running Paired Basecaller Comparisons The most informative validation design runs the same raw signal data through multiple basecallers and compares the outputs. This paired comparison isolates basecaller performance from sequencing run variability. Select a subset of your reference samples representing the range of modification levels you expect in your actual experiments. Process the identical FAST5 or POD5 files through Guppy, Dorado, and Megalodon using the appropriate models for your flowcell and chemistry. The DeepMod2 study demonstrated comparable performance between open-source and closed-source methods when evaluated on the same datasets, providing a template for how such comparisons should be structured. Record the basecaller version, model version, and configuration parameters for each run to ensure the comparison is reproducible. After basecalling, compare the modification calls at three levels. First, compare the overall proportion of modified bases detected by each basecaller. Second, examine per-read modification probabilities to assess consistency. Third, compare modification calls at specific genomic positions to identify systematic differences between basecallers. ### Calculating Agreement Metrics Quantify the agreement between basecallers using position-level comparison. For each genomic position covered by your reference samples, determine whether each basecaller called the base as modified or unmodified. Calculate the concordance rate as the proportion of positions where two basecallers agree. The DeepMod2 study reported a high correlation of 0.96 between reduced representation and whole-genome Nanopore sequencing for methylation detection, illustrating the level of agreement achievable when methods perform consistently. For positions where basecallers disagree, examine the modification probability values. Disagreements with probabilities near the threshold suggest borderline calls that may vary with model parameters. Disagreements with high-confidence probabilities in opposite directions indicate more fundamental differences in how the basecallers interpret the signal. Also compare modification calls against your known reference status. Calculate sensitivity as the proportion of known modified positions correctly identified and specificity as the proportion of known unmodified positions correctly classified. These metrics provide an absolute benchmark that does not depend on comparing one basecaller to another. ### Testing Model and Parameter Sensitivity Basecaller performance depends on model selection and configuration parameters. Test how sensitive your results are to these choices by running the same data with different model versions and parameter settings. The Lokatt study demonstrated that model architecture innovations can achieve performance on par with existing state-of-the-art structures, with a median single read identity score of 0.930 and a genome coverage ratio of 99.750 percent. This finding suggests that model choice materially affects basecalling outcomes. For Guppy and Dorado, test the available model versions for your flowcell type. Compare modification calls across model versions to identify whether newer models produce substantially different results. For Megalodon, test different configuration parameters that affect modification probability estimation. Document the results of these sensitivity tests in your validation records. This documentation helps you understand how robust your modification calls are to technical choices and provides context for interpreting results from different runs. ### Establishing Pass-Fail Criteria Define objective criteria for determining whether your basecaller configuration passes validation. These criteria should be specific to your research questions and modification detection requirements. For a typical methylation study, reasonable criteria include achieving at least 90 percent concordance with known reference status for the positive control and at least 95 percent correct classification of the negative control. Set thresholds based on your downstream analysis requirements. If you plan to compare modification levels between experimental groups, your validation should demonstrate that the basecaller can reliably detect biologically meaningful differences. The Mucor research showed that epimutations can confer antimicrobial resistance through RNAi-dependent silencing or heterochromatin-mediated silencing via H3K9 dimethylation, with distinct stability characteristics during infection. Detecting such biologically significant modification changes requires a basecaller with sufficient sensitivity and specificity. Document your pass-fail criteria before running the validation to avoid bias in interpreting results. If your initial configuration fails validation, investigate the cause before adjusting criteria. Common causes of validation failure include model mismatch, insufficient sequencing depth, and sample quality issues. ### Recording Validation Results Maintain a validation log that records all relevant information for each basecaller configuration tested. Include the date, software versions, model versions, reference samples used, sequencing run identifiers, and the calculated agreement and accuracy metrics. This log serves as evidence that your basecalling pipeline produces reliable modification calls and supports reproducibility across projects. Store the validation log alongside your sequencing data and analysis scripts. The nf-core documentation describes community pipeline standards for reproducible workflow configuration, and following similar principles for validation documentation supports collaboration and long-term data integrity. The Galaxy Training Network offers accessible workflow training that emphasizes reproducibility, providing practical guidance for structuring validation workflows. ### Revalidating After Changes Revalidation is required whenever you change any component of your basecalling pipeline. This includes updating to a new basecaller version, switching to a different model, changing flowcell types, or modifying chemistry protocols. The closed-source nature of Guppy and Dorado means you cannot inspect the underlying changes between versions, making empirical revalidation essential. The DeepMod2 study noted that Guppy and Dorado remain closed-source, which limits independent verification of their underlying algorithms. This limitation increases the importance of your own validation testing when new versions are released. Schedule revalidation as part of your standard workflow when updating software. For projects spanning extended periods, periodically revalidate using stored reference samples to confirm that your pipeline continues to perform consistently. This practice detects drift in basecaller performance or changes in your computational environment that might affect results. ### Troubleshooting Validation Failures When validation fails, work through potential causes systematically. First, confirm that you used the correct model for your flowcell and chemistry. Model mismatch is a common cause of poor modification detection and produces symptoms including low read identity scores and unusual modification proportions. Second, examine sequencing quality metrics for the reference samples. Poor signal quality from degraded DNA or flowcell issues reduces basecalling accuracy and disproportionately affects modification detection because modified base signals are subtle. Check pore occupancy, read length distributions, and basecall quality scores. Third, verify that your reference samples have the expected modification status. The Mucor research demonstrated that epigenetic states can be transient and revert without DNA sequence changes, so reference samples may lose their expected modification patterns over time or with passage. Reconfirm reference sample status using an independent method if validation failures persist. Fourth, review your computational environment for resource exhaustion or configuration issues. Insufficient memory, disk space, or GPU resources can produce incomplete or corrupted output that fails validation. If validation continues to fail after working through these causes, escalate to a bioinformatics specialist or the software provider with your validation records and error documentation. Persistent failures may indicate software bugs or incompatibilities that require expert diagnosis. ## Frequently Asked Questions ### What is the main difference between Guppy and Dorado for modification detection? Guppy and Dorado both provide closed-source modification detection from Oxford Nanopore Technologies, but Dorado represents the current generation with newer model architectures and faster processing. Dorado provides updated models for newer flowcell types and chemistries, while Guppy remains widely supported in established workflows. The DeepMod2 study grouped Guppy and Dorado together as current state-of-the-art methods, indicating comparable performance for methylation detection. ### Can I use Megalodon for high-throughput projects? Megalodon performs signal-level analysis that is computationally intensive, making it impractical for very large datasets. The tool is better suited for smaller projects, targeted analyses, or situations where the open-source nature and per-read modification probabilities justify the additional processing time. For high-throughput projects, Guppy or Dorado with GPU acceleration provide more practical throughput. ### How do I know which basecaller model to use for my flowcell? The correct model depends on your flowcell type and chemistry version. Guppy and Dorado provide model recommendations through their software interfaces and documentation. Using a mismatched model produces poor basecalling accuracy and unreliable modification calls. Verify that the model version matches your sequencing conditions before running the basecaller. ### What output format do I need for downstream methylation analysis? Most methylation analysis tools accept BAM files with MM and ML tags, which Guppy and Dorado produce when configured for modification detection. Megalodon produces per-read modification probabilities that require custom parsing for many tools. Check the documentation for your specific downstream tools to confirm which output formats they support. ### Can I reanalyze my data with a different basecaller later? You can reanalyze data if you retain the raw signal files in FAST5 or POD5 format. Raw signal files allow you to run any basecaller at any time. If you store only basecalled output, your analysis options are limited to what the original basecaller produced. Consider your reanalysis needs when deciding what data to store. ### How does basecaller accuracy affect structural variant detection? Basecaller accuracy directly affects structural variant calling because mis-called bases can create false variant signals. Higher accuracy basecallers reduce false positives in structural variant analysis. The Lokatt study reported a median single read identity score of 0.930 and a genome coverage ratio of 99.750%, providing benchmarks for evaluating basecalling performance in the context of downstream analysis. ### What should I do if my modification calls look unusual? First verify that you used the correct basecaller model for your flowcell and chemistry. Then check signal quality metrics to rule out technical issues. If problems persist, compare your modification levels with expected values from literature or preliminary experiments. Consult with a domain expert if unexpected patterns may indicate biological phenomena of interest, as research on Mucor fungi demonstrated that transient epimutations can produce meaningful modification changes without DNA mutations. ### Is the open-source approach viable for modification detection? The DeepMod2 study demonstrated that open-source frameworks can achieve comparable performance to Guppy and Dorado for methylation detection. Open-source tools provide transparency and customization that closed-source tools lack, but they require more bioinformatics expertise to operate. The viability of an open-source approach depends on your specific requirements for throughput, accuracy, and documentation. ## Related Bioinformatics Guides - [Oxford Nanopore Sequencing: From Sample to Base Calls](/knowledge/bioinformatics/oxford-nanopore-sequencing-from-sample-to-base-calls) - [Single-Cell vs Single-Nucleus RNA Sequencing: Choosing the Right Approach](/knowledge/bioinformatics/single-cell-vs-single-nucleus-rna-sequencing-choosing-the-right-approach) - [How to Choose a Long-Read Sequencing Platform: PacBio vs Oxford Nanopore](/knowledge/bioinformatics/how-to-choose-a-long-read-sequencing-platform-pacbio-vs-oxford-nanopore) - [Metagenomics vs Metabarcoding: Choosing the Right Approach for Your Study](/knowledge/bioinformatics/metagenomics-vs-metabarcoding-choosing-the-right-approach-for-your-study) - [Single-Cell Sequencing Services: How to Choose a Provider](/knowledge/bioinformatics/single-cell-sequencing-services-how-to-choose-a-provider) ## References and Further Reading - [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/). National Center for Biotechnology Information. - [EMBL-EBI Training](https://www.ebi.ac.uk/training). European Bioinformatics Institute. - [Bioconductor](https://bioconductor.org/). Bioconductor Project. - [Galaxy Training Network](https://training.galaxyproject.org/). Galaxy Project. - [nf-core Documentation](https://nf-co.re/docs). nf-core. - [The Carpentries Lessons](https://carpentries.org/lessons). The Carpentries. - [A signal processing and deep learning framework for methylation detection using Oxford Nanopore sequencing.](https://doi.org/10.1038/s41467-024-45778-y). 2024. - [Lokatt: a hybrid DNA nanopore basecaller with an explicit duration hidden Markov model and a residual LSTM network.](https://doi.org/10.1186/s12859-023-05580-x). 2023. - [Epimutations driven by RNAi or heterochromatin evoke transient antimicrobial drug resistance in pathogenic Mucor fungi.](https://doi.org/10.1371/journal.pbio.3003598). 2026. > This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.