# Mutational Signature Analysis in Somatic Variant Calling: From COSMIC Signatures to Clinical Insights ## Key Takeaways - Mutational signature analysis fundamentally relies on the accuracy of preceding somatic variant calling, as subtle shifts in mutation type frequencies, even from systematic artifacts like excess C-to-T transitions, can significantly distort inferred signature exposures and lead to erroneous conclusions about underlying mutagenic processes. - The distinction between somatic and germline variants is critical; accurate somatic variant calling necessitates high-quality, matched normal samples to differentiate true somatic mutations from inherited polymorphisms and sequencing artifacts, with low-depth or contaminated normal samples posing significant risks of misclassification. - The COSMIC framework provides a standardized reference collection of mutational signatures, enabling cross-study comparability by defining signatures based on trinucleotide contexts, but its limitations include potential incompleteness and the need to be aware of specific catalog versions due to ongoing refinement and discovery of new signatures. - Both reference-based signature assignment and de novo signature discovery are viable strategies, with assignment being suitable for hypothesis testing and comparability to existing literature, while de novo discovery is powerful for characterizing poorly understood mutational landscapes or identifying novel signatures, especially with large sample sizes. - Reproducibility in mutational signature analysis is paramount, requiring meticulous documentation of all workflow components, including reference genome versions, variant callers, filtering parameters, signature catalog versions, and analysis tools, to ensure consistent and verifiable results. - Interpretation of mutational signatures requires careful consideration of reconstruction error, which quantifies how well assigned signatures explain the observed mutation catalogue, and confidence intervals for signature exposures, particularly in non-cancer tissues where etiological links are less established than in cancer. --- Somatic variant calling identifies mutations present in tumor or tissue samples that are absent from the germline, and mutational signature analysis interprets the patterns of those mutations to infer the biological processes that generated them. This article explains how to move from raw sequencing data through somatic variant calling to mutational signature extraction and interpretation, with attention to workflow choices, quality controls, reproducibility, and the clinical and biological limits of the approach. The intended readers are biology students, researchers, laboratory professionals, and life-science practitioners who need a practical framework for conducting this analysis and for judging when results are reliable enough to inform further investigation. ## The Analytical Pipeline from Sequencing Reads to Mutational Signatures Mutational signature analysis depends entirely on the quality and accuracy of the somatic variant calls that precede it. A mutational signature is a pattern of mutation types, typically defined by the six classes of single base substitutions and their immediate 5-prime and 3-prime sequence context, which yields 96 possible trinucleotide contexts. The underlying assumption, supported by large-scale cancer genomics, is that the catalogue of somatic mutations in a cancer genome carries the imprint of the mutational processes that were active during tumor development. The original extraction of more than 20 distinct signatures from nearly five million mutations across over seven thousand cancers demonstrated that some signatures recur across many cancer types while others are confined to specific cancer classes, and that certain signatures associate with patient age, known mutagenic exposures, or defects in DNA maintenance [1]. The practical workflow proceeds through several stages. First, sequencing reads are aligned to a reference genome. Second, somatic variants are called by comparing tumor and matched normal samples. Third, variants are filtered to remove artifacts and likely germline polymorphisms. Fourth, the remaining somatic mutations are classified into mutation types and trinucleotide contexts. Fifth, the observed mutation catalogue is decomposed into a weighted combination of known signatures or used to discover novel signatures. Sixth, the resulting signature exposures are interpreted in the context of the sample's clinical and biological features. Each stage introduces potential errors that propagate downstream. A variant calling error rate that is acceptable for identifying driver mutations may be unacceptable for signature analysis, because signatures are sensitive to the relative frequencies of mutation types. A systematic artifact that produces excess C-to-T transitions, for example, will distort the inferred signature exposures and may lead to false conclusions about the involvement of specific mutational processes. The sensitivity of modern sequencing methods has expanded the scope of somatic mutation detection beyond clonal populations. A duplex sequencing protocol with error rates below five errors per billion base pairs enabled the detection of somatic mutations in single DNA molecules from cell populations, allowing the study of somatic mutagenesis in any tissue independently of clonality [2]. This technical advance demonstrates that the quality of the underlying mutation detection directly determines the reliability of downstream signature analysis, particularly in tissues where mutations are present at low allele frequencies. ## Somatic versus Germline Variant Calling and the Importance of Matched Normals The distinction between somatic and germline variant calling is central to mutational signature analysis. Germline variants are inherited polymorphisms present in every cell of an individual, while somatic mutations arise during the lifetime of the organism and are present only in subsets of cells. Somatic variant calling requires a matched normal sample from the same individual to distinguish true somatic mutations from germline polymorphisms and from sequencing artifacts. The standard design uses a tumor sample and a matched normal sample, usually from blood or adjacent non-malignant tissue. The variant caller compares the two samples and reports variants that are present in the tumor but absent from the normal. This design controls for the individual's germline background and reduces the risk of mistaking common polymorphisms for somatic mutations. When a matched normal is unavailable, somatic variant calling becomes substantially more difficult, and the resulting mutation catalogue will contain germline contamination that distorts signature analysis. The quality of the normal sample matters as much as the quality of the tumor sample. A low-depth normal sample may fail to detect a germline variant that is actually present, causing that variant to be misclassified as somatic. A contaminated normal sample, containing tumor cells, may lead to the opposite error, causing true somatic mutations to be filtered out. Laboratory professionals should verify the depth and purity of both tumor and normal samples before proceeding with variant calling. Variant filtering is the step where many practical errors are caught or missed. Filters typically address read depth, variant allele frequency, mapping quality, strand bias, and proximity to known artifact-prone regions. The goal is to retain true somatic mutations while removing sequencing errors and alignment artifacts. The filtering strategy should be documented and applied consistently across all samples in a study, because inconsistent filtering will produce systematic differences in mutation catalogs that can be misinterpreted as biological differences. The challenge of distinguishing true somatic mutations from artifacts is particularly acute in studies of normal tissues and age-related diseases. A study of somatic mutations in human chondrocytes used single-cell whole-genome sequencing to analyze single-nucleotide variants and small insertions and deletions in cells isolated from cartilage, demonstrating that both mutation types accumulate with age [3]. The success of this analysis depended on rigorous variant calling that could distinguish genuine somatic mutations from the artifacts introduced during whole-genome amplification of single cells. ## The COSMIC Signature Framework and Its Role in Interpretation The Catalogue of Somatic Mutations in Cancer, maintained by the Wellcome Sanger Institute and accessible through the [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/) ecosystem, provides the reference collection of mutational signatures used by most analysis tools. The COSMIC framework organizes signatures by mutation type, with single base substitution signatures, doublet base substitution signatures, and small insertion and deletion signatures. Each signature is defined by a probability distribution over mutation types, and a tumor's mutation catalogue is modeled as a weighted sum of these signature distributions. The original description of mutational signatures in human cancer established the conceptual foundation for this framework. The analysis of thousands of cancer genomes revealed that some signatures are ubiquitous, such as those attributed to the APOBEC family of cytidine deaminases, while others are restricted to particular cancer types. Some signatures correlate with patient age at diagnosis, known mutagenic exposures such as tobacco smoke or ultraviolet light, or defects in DNA repair pathways. Many signatures remain of unknown origin, which limits their interpretive value [1]. The COSMIC framework is useful because it provides a common language for describing mutational processes across studies and cancer types. A researcher who identifies a dominant signature in a tumor sample can compare that result with published data from other cohorts and with experimental studies that have linked specific signatures to specific exposures or repair defects. This comparability is the main strength of the reference-based approach. The framework also has limitations. The reference signatures were derived from specific cohorts and may not capture all mutational processes operating in human tissues. The original extraction identified more than 20 signatures, but subsequent analyses with improved methods have expanded the catalog and refined the definitions of existing signatures. A reanalysis of more than 2,700 cancer genomes with the MuSiCal framework produced an improved catalog of signatures, discovered nine indel signatures absent from the current catalog, and resolved long-standing issues with ambiguous flat signatures [4]. This finding illustrates that the reference catalog is not static and that researchers should be aware of the version of the catalog they are using. The application of the COSMIC framework to non-cancer tissues requires additional caution. Studies of normal tissues and age-related diseases have identified clock-like signatures that accumulate with age, but the interpretation of these signatures in terms of specific biological processes is less well established than in cancer. A study of osteoarthritis cartilage found differences in mutational signatures between affected and unaffected samples, but the clinical significance of these differences requires further investigation [3]. ## Reference-Based Signature Assignment versus De Novo Signature Discovery Two broad analytical strategies exist for mutational signature analysis. Reference-based assignment decomposes an observed mutation catalogue using a predefined set of signatures. De novo discovery extracts signatures directly from the data without reference to an existing catalog. Each strategy has distinct requirements, strengths, and failure modes. Reference-based assignment is appropriate when the researcher has a hypothesis about which mutational processes are active or when the sample size is too small to support de novo discovery. The approach is computationally efficient and produces results that are directly comparable with published data. The main risk is that the reference catalog may not contain the true signatures operating in the sample, in which case the assignment algorithm will produce a weighted combination of available signatures that approximates the observed data but does not reflect the underlying biology. De novo discovery is appropriate when the researcher is studying a tissue or disease for which the mutational process landscape is poorly characterized or when the sample size is large enough to support statistical extraction. The approach can reveal novel signatures and provide a more accurate picture of the mutational processes operating in the sample. The main risks are overfitting, instability of the extracted signatures, and difficulty in matching discovered signatures to known biological processes. The choice between the two strategies should be made before analysis and documented in the study protocol. A common practical approach is to perform de novo discovery on a large discovery cohort, validate the discovered signatures in an independent cohort, and then use reference-based assignment for individual samples. This hybrid approach balances the strengths of both strategies. The MuSiCal framework represents a recent advance in both discovery and assignment. Its simulation studies demonstrated improved performance over state-of-the-art algorithms for both tasks, and its reanalysis of thousands of cancer genomes produced an improved signature catalog [4]. Researchers who are establishing a mutational signature analysis pipeline should consider whether their chosen tools incorporate these methodological improvements. A study of chronic lymphocytic leukemia demonstrated the value of de novo signature discovery in a clinical context. The analysis of 209 patient samples identified a prominent endogenous mutational process leading to C-to-T single base substitutions, and the de novo extraction revealed two novel single base substitution signatures associated with immunoglobulin heavy chain variable region hypermutated and unmutated status [5]. This finding illustrates how de novo discovery can reveal disease heterogeneity that is not captured by traditional risk stratification methods. ## Practical Workflow for Mutational Signature Analysis The following workflow describes the steps required to move from somatic variant calls to interpretable mutational signature results. The workflow assumes that somatic variant calling has already been completed and that the researcher has a VCF file or similar variant format containing somatic mutations. ### Step 1: Prepare the Mutation Catalogue The first step is to convert the somatic variant calls into a mutation catalogue formatted for signature analysis. This requires extracting the chromosome, position, reference allele, and alternate allele for each variant, and then classifying each variant by mutation type and trinucleotide context. The trinucleotide context is the reference sequence surrounding the mutated base, typically one base upstream and one base downstream. The mutation classification must account for strand symmetry. A C-to-T mutation on the forward strand is equivalent to a G-to-A mutation on the reverse strand, because the reference genome is double-stranded. Signature analysis tools typically normalize mutations to the pyrimidine base, so that all mutations are represented as C-to-X or T-to-X changes. This normalization is essential for comparing mutation patterns across samples and for matching observed patterns to reference signatures. ### Step 2: Select the Analysis Tool Several tools are available for mutational signature analysis, and the choice of tool affects the results. The [Bioconductor](https://bioconductor.org/) project hosts multiple R packages for signature analysis, including packages for signature fitting, signature discovery, and visualization. These packages are documented and versioned, which supports reproducible analysis. The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible tutorials for mutational signature analysis that can be run without command-line expertise. These tutorials are useful for researchers who are new to the field and for laboratory professionals who need to understand the analysis steps before implementing them in their own environment. The [nf-core](https://nf-co.re/docs) community provides standardized pipelines for genomic analysis, including pipelines that incorporate somatic variant calling and mutational signature analysis. These pipelines follow community standards for configuration, usage, and reproducibility, which is valuable for laboratories that need consistent results across many samples. ### Step 3: Perform Signature Assignment or Discovery The core analysis step is either assigning the observed mutation catalogue to known signatures or discovering novel signatures from the data. Reference-based assignment tools typically require the mutation catalogue and a signature reference file as input, and they output the estimated weights of each signature in the sample. De novo discovery tools require a collection of mutation catalogues from multiple samples and output a set of discovered signatures and their weights in each sample. The number of mutations in the sample is a critical constraint. Signature analysis requires a minimum number of mutations to produce reliable results, and samples with very low mutation burdens may not be amenable to analysis. The exact minimum depends on the tool and the number of signatures being considered, but researchers should be cautious when interpreting signature exposures from samples with fewer than a few hundred somatic mutations. ### Step 4: Validate and Interpret the Results The final step is to validate the results and interpret them in the biological and clinical context of the sample. Validation involves checking that the reconstructed mutation catalogue, computed as the weighted sum of the assigned signatures, closely matches the observed mutation catalogue. Poor reconstruction indicates that the reference signatures do not adequately explain the observed mutations. Interpretation requires knowledge of the biological processes associated with each signature. Some signatures have well-established etiologies, such as the clock-like signatures associated with age, the APOBEC signatures associated with cytidine deaminase activity, and the signatures associated with specific DNA repair defects. Other signatures remain of unknown origin, and their presence in a sample should be reported without overinterpretation. ## Tools and Resources for Reproducible Analysis Reproducibility is a central concern in mutational signature analysis because the results depend on many analytical choices, including the reference genome version, the variant caller, the filtering thresholds, the signature catalog version, and the analysis tool. A reproducible analysis documents all of these choices and makes the analysis code and input data available for verification. The [EMBL-EBI Training](https://www.ebi.ac.uk/training) program offers learning pathways for bioinformatics that cover data resources, analysis methods, and practical skills. These training materials are useful for researchers who need to build the computational skills required for mutational signature analysis, including working with genomic data formats, using command-line tools, and writing analysis scripts. The [Carpentries](https://carpentries.org/lessons) lessons provide foundational training in computing, data analysis, shell scripting, Git version control, and programming. These skills are essential for conducting reproducible mutational signature analysis, because they enable researchers to organize their workflows, track changes to their analysis code, and share their methods with others. The [Galaxy Training Network](https://training.galaxyproject.org/) provides workflow-based training that emphasizes reproducibility through the use of shared workflows and histories. Galaxy workflows can be saved, versioned, and shared, which makes them a practical option for laboratories that need to run the same analysis repeatedly on different samples. The [nf-core](https://nf-co.re/docs) documentation describes community standards for pipeline development and usage. Pipelines that follow these standards include configuration files, usage documentation, and containerized environments, which support reproducible execution across different computing infrastructures. A practical reproducibility checklist for mutational signature analysis includes the following items. Record the reference genome version and source. Record the variant caller version and all filtering parameters. Record the signature catalog version and source. Record the analysis tool version and all parameters. Save the mutation catalogue in a structured format. Save the signature assignment or discovery results. Document the interpretation and any limitations of the analysis. ## Observations and Measurements That Support Interpretation Mutational signature analysis produces quantitative outputs that can be used to compare samples and to correlate with clinical and biological features. The primary output is the signature exposure, which is the estimated contribution of each signature to the sample's mutation burden. Signature exposures can be expressed as the number of mutations attributed to each signature or as the fraction of total mutations attributed to each signature. The total mutation burden is an important measurement that provides context for signature interpretation. A sample with a high mutation burden may have sufficient mutations for reliable signature analysis, while a sample with a low mutation burden may not. The mutation burden also has biological significance, because different mutational processes produce different numbers of mutations. The reconstruction error, which measures how well the assigned signatures explain the observed mutations, is a key quality metric. A high reconstruction error indicates that the reference signatures do not adequately explain the observed data, which may reflect the presence of novel signatures or the effects of sequencing artifacts. The stability of the signature assignment is another useful measurement. Some tools provide confidence intervals or bootstrap estimates for signature exposures, which indicate how sensitive the results are to sampling variation. Unstable assignments, with wide confidence intervals, should be interpreted with caution. The single-cell and single-molecule studies of somatic mutation have expanded the scope of mutational signature analysis beyond cancer. A study using duplex sequencing with error rates below five errors per billion base pairs demonstrated that somatic mutations can be detected in single DNA molecules from cell populations, enabling the study of somatic mutagenesis in any tissue independently of clonality. This study found that differentiated cells in blood and colon had mutation loads and signatures similar to their corresponding stem cells, and that post-mitotic neurons accumulate somatic mutations at a constant rate throughout life. These findings indicate that mutational processes independent of cell division are important contributors to somatic mutagenesis [2]. A single-cell study of chondrocytes from human cartilage found that both single-nucleotide variants and small insertions and deletions accumulate with age, with a clock-like mutational signature. The study compared chondrocytes from patients with osteoarthritis and non-osteoarthritis donors and found differences in mutational signatures between the two groups. This work demonstrates the application of mutational signature analysis to age-related diseases beyond cancer [3]. A study of chronic lymphocytic leukemia examined mutational signatures in 209 patient samples and identified a prominent endogenous mutational process leading to C-to-T single base substitutions. The study identified two novel de novo single base substitution signatures associated with immunoglobulin heavy chain variable region hypermutated and unmutated status, and found that a notable fraction of the cohort exhibited a signature suggesting a link to occupational haloalkane exposure. This work illustrates how mutational signature analysis can reveal potential environmental risk factors and disease heterogeneity that traditional risk stratification methods may miss [5]. ## Common Failure Patterns and How to Avoid Them Several recurring problems undermine mutational signature analysis. Recognizing these failure patterns is essential for producing reliable results. The first failure pattern is germline contamination. When the normal sample is absent, low quality, or contaminated, germline polymorphisms are misclassified as somatic mutations. Because germline polymorphisms have a different mutation pattern than true somatic mutations, this contamination distorts the signature analysis. The solution is to use high-quality matched normal samples and to apply rigorous variant filtering. The second failure pattern is artifact-driven mutation patterns. Sequencing artifacts, such as oxidative damage during library preparation or errors introduced during PCR amplification, produce characteristic mutation patterns that can be mistaken for biological signatures. The solution is to use library preparation methods that minimize artifacts, to apply strand bias filters, and to compare mutation patterns between technical replicates. The third failure pattern is overinterpretation of low-mutation samples. When a sample has few somatic mutations, the signature assignment is based on limited data and is highly uncertain. The solution is to report confidence intervals for signature exposures and to avoid drawing strong conclusions from samples with low mutation burdens. The fourth failure pattern is using an outdated signature catalog. The COSMIC catalog has been updated multiple times, and newer versions include additional signatures and refined definitions of existing signatures. The solution is to record the catalog version used in the analysis and to check whether published comparisons use the same version. The fifth failure pattern is ignoring reconstruction error. When the assigned signatures do not reconstruct the observed mutation catalogue well, the results are unreliable. The solution is to always compute and report the reconstruction error and to investigate samples with high error. The sixth failure pattern is circular interpretation. When a researcher assigns signatures using a reference catalog and then interprets the results in terms of the same catalog, the interpretation is circular. The solution is to validate signature assignments using independent data, such as experimental exposure studies or orthogonal sequencing methods. ## Limitations of Mutational Signature Analysis Mutational signature analysis has inherent limitations that should be acknowledged in any report of results. The most fundamental limitation is that signatures are statistical patterns, not direct measurements of biological processes. A signature attributed to a specific exposure or repair defect is a hypothesis about the underlying biology, not a proof. The reference signature catalog is incomplete. Many signatures in the catalog have unknown etiologies, and the catalog may not contain signatures for all mutational processes operating in human tissues. The discovery of new signatures in recent analyses demonstrates that the catalog continues to evolve [4]. Signature assignment is non-unique. Different combinations of signatures can produce similar reconstructed mutation catalogues, particularly when signatures have similar mutation patterns. This non-uniqueness means that the estimated signature exposures are not necessarily the true exposures. The resolution of signature analysis is limited by the mutation burden of the sample. Samples with low mutation burdens cannot support the estimation of many signature exposures, and the results are correspondingly uncertain. The interpretation of signatures in non-cancer tissues is less well established than in cancer. While studies have demonstrated the presence of clock-like signatures in normal tissues and in age-related diseases, the clinical significance of these signatures is still being investigated [3]. The clinical utility of mutational signature analysis is an active area of research. Some studies have demonstrated associations between specific signatures and prognosis or treatment response, but these associations require validation in prospective cohorts before they can inform clinical decisions. The study of chronic lymphocytic leukemia identified associations between novel signatures and disease heterogeneity, but the clinical implications of these findings require further investigation [5]. ## Safety and Regulatory Context for Clinical Applications Researchers who plan to use mutational signature analysis in clinical contexts should be aware of the regulatory and ethical considerations that apply to genomic analysis. The use of patient samples requires appropriate ethical approval and informed consent, and the handling of genomic data must comply with applicable privacy regulations. The reporting of mutational signature results to patients or clinicians requires careful consideration of the uncertainty and limitations of the analysis. Signature exposures are estimates with associated uncertainty, and the biological interpretation of signatures is often incomplete. Results should be reported with appropriate caveats and should not be presented as definitive diagnoses or prognostic indicators unless they have been validated for that purpose. Laboratory professionals who develop mutational signature analysis pipelines for clinical use should validate their pipelines using reference samples with known mutation patterns. Validation should demonstrate that the pipeline produces accurate and reproducible results across runs and operators. Professional escalation criteria for mutational signature analysis include the following situations. Escalate when the reconstruction error is high and cannot be reduced by adjusting the analysis parameters. Escalate when the signature assignment is unstable across repeated analyses. Escalate when the results suggest a novel signature that could have clinical implications. Escalate when the results conflict with other clinical or laboratory findings. Escalate when the analysis is being used to inform a clinical decision and the limitations of the analysis have not been adequately addressed. ## At a Glance | Analysis Stage | Key Decision | Common Error | Quality Control | | --- | --- | --- | --- | | Variant calling | Use matched normal sample | Germline contamination | Verify normal sample purity and depth | | Variant filtering | Apply consistent filters across samples | Inconsistent filtering creates false biological differences | Document all filter parameters and thresholds | | Mutation classification | Normalize to pyrimidine base | Strand asymmetry distorts mutation patterns | Verify trinucleotide context extraction | | Signature assignment | Choose reference catalog version | Outdated catalog misses novel signatures | Record catalog version and source | | Signature discovery | Require sufficient sample size | Overfitting produces unstable signatures | Validate discovered signatures in independent cohort | | Result interpretation | Match signatures to known etiologies | Overinterpretation of unknown signatures | Report reconstruction error and confidence intervals | ## Records and Measurements for Quality Assurance Maintaining detailed records is essential for reproducible mutational signature analysis. The following records should be kept for each analysis run. The sample manifest should record the sample identifier, tissue type, sequencing platform, coverage, and the path to the tumor and normal sequencing data. This manifest provides the context for interpreting the analysis results. The variant calling record should document the variant caller version, the reference genome version, the alignment tool and parameters, and the filtering thresholds applied. This record enables the analysis to be reproduced or modified. The mutation catalogue should be saved in a structured format that includes the chromosome, position, reference allele, alternate allele, mutation type, and trinucleotide context for each variant. This file is the input to the signature analysis and should be archived. The signature analysis record should document the analysis tool version, the signature catalog version, the parameters used, and the output files. The output files should include the signature exposures, the reconstruction error, and any confidence intervals. The interpretation record should document the biological and clinical context for the interpretation, the evidence supporting the assignment of signatures to specific etiologies, and any limitations of the interpretation. The following measurements should be reported for each sample. The total number of somatic mutations. The number of mutations in each of the six single base substitution classes. The number of mutations in each trinucleotide context. The estimated exposure of each assigned signature. The reconstruction error. The confidence intervals for the signature exposures. ## Decision Framework for Selecting Between Reference-Based Assignment and De Novo Discovery Choosing between reference-based signature assignment and de novo discovery is a consequential decision that shapes the entire downstream analysis. The choice depends on the research question, the sample size, the tissue type, and the maturity of the mutational process landscape for the condition under study. A structured decision framework helps researchers make this choice deliberately instead of by default. ### Decision Criteria for Reference-Based Assignment Reference-based assignment is the appropriate choice when the researcher has a specific hypothesis about which mutational processes are active in the samples. This approach works well when the study aims to test whether known mutational processes, such as those associated with age, tobacco exposure, or specific DNA repair defects, are present in a new cohort. The approach is also suitable when the sample size is small, because reference-based assignment requires fewer samples than de novo discovery to produce stable results. The practical threshold for reference-based assignment depends on the number of signatures being considered and the mutation burden of each sample. A sample with fewer than a few hundred somatic mutations will produce uncertain signature exposures regardless of the assignment method. Researchers should evaluate whether their samples meet this minimum before proceeding with either approach. Reference-based assignment is also appropriate when the researcher needs results that are directly comparable with published data. Because the reference catalog provides a common language for describing mutational processes, assignment results can be compared across studies that use the same catalog version. This comparability is valuable for meta-analyses and for validating findings in independent cohorts. The main limitation of reference-based assignment is that it cannot detect mutational processes that are absent from the reference catalog. If the samples contain a novel mutational process, the assignment algorithm will produce a weighted combination of available signatures that approximates the observed data but does not reflect the underlying biology. This limitation is particularly relevant for studies of tissues or diseases that are underrepresented in the reference catalog. ### Decision Criteria for De Novo Discovery De novo discovery is the appropriate choice when the researcher is studying a tissue or disease for which the mutational process landscape is poorly characterized. This approach is also suitable when the study aims to identify novel mutational processes or to refine the definitions of existing signatures. De novo discovery requires a larger sample size than reference-based assignment, because the extraction algorithm needs sufficient statistical power to separate distinct mutational processes. The sample size requirement for de novo discovery depends on the number of distinct signatures operating in the samples and the similarity of their mutation patterns. Signatures with similar patterns are harder to separate and require more samples. Researchers should plan for a discovery cohort of at least several dozen samples, and ideally more, to support reliable de novo extraction. The exact number depends on the mutation burden of each sample and the complexity of the underlying mutational process landscape. De novo discovery is also appropriate when the researcher suspects that the reference catalog is incomplete for the condition under study. A study of chronic lymphocytic leukemia used de novo extraction to identify two novel single base substitution signatures that were not present in the reference catalog, and these signatures correlated with immunoglobulin heavy chain variable region hypermutated and unmutated status [5]. This finding demonstrates that de novo discovery can reveal disease heterogeneity that reference-based assignment would miss. The main limitation of de novo discovery is the risk of overfitting. When the sample size is too small or the number of extracted signatures is too large, the algorithm may produce signatures that fit the observed data well but do not reflect real biological processes. Validation in an independent cohort is essential to distinguish genuine signatures from statistical artifacts. ### A Practical Decision Matrix The following decision matrix summarizes the key considerations for choosing between the two approaches. | Consideration | Reference-Based Assignment | De Novo Discovery | | --- | --- | --- | | Research question | Test known mutational processes | Discover novel mutational processes | | Sample size | Small cohorts, individual samples | Large cohorts, dozens or more samples | | Tissue or disease maturity | Well-characterized mutational landscape | Poorly characterized mutational landscape | | Comparability with published data | Directly comparable | Requires matching to known signatures | | Risk of missing novel processes | High | Low | | Risk of overfitting | Low | High | | Validation requirement | Moderate | High, requires independent cohort | ### Hybrid Approach for Comprehensive Analysis A hybrid approach that combines both strategies is often the most informative. The hybrid approach proceeds in three stages. First, perform de novo discovery on a large discovery cohort to identify the mutational processes operating in the samples. Second, validate the discovered signatures in an independent cohort to confirm that they are reproducible and not artifacts of the discovery cohort. Third, use reference-based assignment with the validated signature set to analyze individual samples, including samples with lower mutation burdens that cannot support de novo extraction. The hybrid approach balances the strengths of both strategies. De novo discovery provides a complete picture of the mutational processes operating in the samples, including novel processes that are absent from the reference catalog. Reference-based assignment provides stable and comparable results for individual samples. The hybrid approach also provides a natural framework for validation, because the discovered signatures are tested in an independent cohort before being used for assignment. The MuSiCal framework represents a recent advance that supports both discovery and assignment within a single analytical framework. Its simulation studies demonstrated improved performance over state-of-the-art algorithms for both tasks, and its reanalysis of more than 2,700 cancer genomes produced an improved catalog of signatures, discovered nine indel signatures absent from the current catalog, and resolved long-standing issues with ambiguous flat signatures [4]. Researchers who are establishing a mutational signature analysis pipeline should consider whether their chosen tools incorporate these methodological improvements. ### Implementation Steps for the Decision Framework The following steps provide a practical procedure for applying the decision framework to a specific study. First, define the research question and the primary hypothesis. If the study aims to test whether known mutational processes are present in a new cohort, reference-based assignment is appropriate. If the study aims to characterize the mutational process landscape of a poorly understood tissue or disease, de novo discovery is appropriate. Second, assess the sample size and mutation burden. Count the number of samples and the number of somatic mutations per sample. If the cohort has fewer than several dozen samples or if individual samples have fewer than a few hundred mutations, reference-based assignment is the more reliable choice. Third, review the existing literature for the tissue or disease under study. Determine whether mutational signature analyses have been published for this condition and whether the reference catalog is likely to contain the relevant signatures. If the literature is sparse or the reference catalog is likely incomplete, de novo discovery is warranted. Fourth, document the decision and the rationale in the study protocol. The decision should be made before the analysis begins, and the rationale should be recorded so that the analysis can be reproduced and evaluated by others. Fifth, plan for validation. Regardless of the chosen approach, the results should be validated using independent data. For reference-based assignment, validation involves checking the reconstruction error and comparing the results with published data. For de novo discovery, validation involves confirming the discovered signatures in an independent cohort. ### Common Failure Patterns in Method Selection Several recurring problems arise from poor method selection. Recognizing these failure patterns helps researchers avoid them. The first failure pattern is using reference-based assignment when the reference catalog is incomplete for the condition under study. This produces results that appear reliable but miss the true mutational processes operating in the samples. The solution is to review the literature and assess whether the reference catalog is likely to contain the relevant signatures before choosing the assignment approach. The second failure pattern is using de novo discovery with too few samples. This produces unstable signatures that do not reproduce in independent cohorts. The solution is to ensure that the discovery cohort is large enough to support reliable extraction and to validate the discovered signatures in an independent cohort. The third failure pattern is failing to document the method selection rationale. This makes it difficult for others to evaluate the analysis and to determine whether the chosen approach was appropriate for the research question. The solution is to document the decision and the rationale in the study protocol before the analysis begins. The fourth failure pattern is ignoring the mutation burden of individual samples. Even when the cohort is large enough for de novo discovery, individual samples with low mutation burdens may not support reliable signature assignment. The solution is to report the mutation burden for each sample and to interpret signature exposures with appropriate caution for low-mutation samples. ### Records and Measurements for Method Selection The following records should be kept to support the method selection decision and to enable evaluation of the analysis. The study protocol should record the research question, the primary hypothesis, and the rationale for the chosen analysis approach. The protocol should also record the sample size, the expected mutation burden, and the review of the existing literature that informed the decision. The sample manifest should record the number of samples, the tissue type, the sequencing platform, and the mutation burden for each sample. This manifest provides the context for evaluating whether the chosen approach was appropriate. The analysis record should document the analysis tool version, the signature catalog version, the parameters used, and the output files. This record enables the analysis to be reproduced and evaluated by others. The validation record should document the validation strategy, the independent cohort used for validation, and the results of the validation. This record is essential for de novo discovery, where validation in an independent cohort is required to distinguish genuine signatures from statistical artifacts. ### Professional Escalation Criteria for Method Selection The following situations warrant escalation to a supervisor, collaborator, or methodological expert. Escalate when the research question is not clearly defined and the choice between reference-based assignment and de novo discovery cannot be justified. Escalate when the sample size is borderline and the reliability of the chosen approach is uncertain. Escalate when de novo discovery produces signatures that cannot be matched to known biological processes and the interpretation is unclear. Escalate when the results of the analysis will inform a clinical decision and the limitations of the chosen approach have not been adequately addressed. The [EMBL-EBI Training](https://www.ebi.ac.uk/training) program offers learning pathways that cover mutational signature analysis methods and data resources. The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible tutorials that demonstrate both reference-based assignment and de novo discovery workflows. The [Bioconductor](https://bioconductor.org/) project hosts documented R packages for both approaches. The [nf-core](https://nf-co.re/docs) community provides standardized pipelines that incorporate both strategies. The [Carpentries](https://carpentries.org/lessons) lessons provide foundational computing skills that support the implementation of either approach in a reproducible manner. ## Frequently Asked Questions ### What is the minimum number of somatic mutations needed for mutational signature analysis? The minimum number depends on the analysis tool and the number of signatures being considered, but reliable results generally require several hundred mutations. Samples with fewer mutations produce uncertain signature exposures with wide confidence intervals. Researchers should report the mutation burden and the associated uncertainty when interpreting signature results from low-mutation samples. ### How does a matched normal sample improve somatic variant calling? A matched normal sample from the same individual allows the variant caller to distinguish true somatic mutations from germline polymorphisms. Variants present in the tumor but absent from the normal are candidate somatic mutations, while variants present in both samples are germline. Without a matched normal, germline polymorphisms contaminate the somatic mutation catalogue and distort the signature analysis. ### What is the difference between reference-based signature assignment and de novo signature discovery? Reference-based assignment decomposes an observed mutation catalogue using a predefined set of known signatures. De novo discovery extracts signatures directly from the data without reference to an existing catalog. Reference-based assignment is appropriate for hypothesis-driven analysis and small sample sizes, while de novo discovery is appropriate for characterizing poorly understood tissues or diseases and requires larger sample sizes. ### How should I choose between different mutational signature analysis tools? The choice of tool should be based on the analysis goals, the sample size, and the need for reproducibility. Tools available through [Bioconductor](https://bioconductor.org/) provide documented and versioned R packages for signature analysis. The [Galaxy Training Network](https://training.galaxyproject.org/) offers accessible tutorials for researchers without command-line expertise. The [nf-core](https://nf-co.re/docs) community provides standardized pipelines that follow reproducibility standards. ### What does reconstruction error tell me about the quality of my signature assignment? Reconstruction error measures how well the weighted sum of the assigned signatures reproduces the observed mutation catalogue. Low reconstruction error indicates that the reference signatures adequately explain the observed mutations. High reconstruction error indicates that the reference signatures do not capture the mutational processes operating in the sample, which may reflect novel signatures or sequencing artifacts. ### Can mutational signature analysis be applied to non-cancer tissues? Yes. Studies using single-cell and single-molecule sequencing have demonstrated that somatic mutations accumulate in normal tissues and in age-related diseases. A study of chondrocytes found that mutations accumulate with age with a clock-like signature, and a study of osteoarthritis cartilage found differences in mutational signatures between affected and unaffected samples [3]. These applications are less well established than cancer applications but are an active area of research. ### How do I know if a mutational signature has a known cause? The COSMIC catalog and the published literature associate some signatures with specific etiologies, such as age, tobacco smoke, ultraviolet light, APOBEC activity, or DNA repair defects. Many signatures remain of unknown origin. Researchers should consult the current literature and the signature catalog documentation to determine the evidence supporting a specific etiology for each signature. ### What should I do if my sample shows a novel mutational signature? A novel signature should be validated before it is interpreted. Validation involves confirming that the signature is reproducible across samples, that it is not an artifact of the sequencing or analysis methods, and that it is not an artifact of the reference catalog version. The discovery of novel signatures should be reported with appropriate caveats and should be investigated using independent data. ## Related Bioinformatics Guides - [Genomic Data Analysis Tools: A Comparative Guide for Researchers](/knowledge/bioinformatics/genomic-data-analysis-tools-a-comparative-guide-for-researchers) - [How to Interpret Gene Set Enrichment Analysis Results](/knowledge/bioinformatics/how-to-interpret-gene-set-enrichment-analysis-results) - [Metagenome Assembled Genome Analysis: From Bins to Biological Insights](/knowledge/bioinformatics/metagenome-assembled-genome-analysis-from-bins-to-biological-insights) - [Metagenomics Data Analysis: From Raw Reads to Biological Insights](/knowledge/bioinformatics/metagenomics-data-analysis-from-raw-reads-to-biological-insights) - [Proteomics Data Analysis Workflow: From Raw Spectra to Biological Insights](/knowledge/bioinformatics/proteomics-data-analysis-workflow-from-raw-spectra-to-biological-insights) ## Related Clinical & Scientific Guides * [A Practical Guide to Detecting Antimicrobial Resistance Genes in Shotgun Metagenomic Data](/knowledge/bioinformatics/a-practical-guide-to-detecting-antimicrobial-resistance-genes-in-shotgun-metagenomic-data) * [Computational Immunology: Modeling the Immune System](/knowledge/bioinformatics/computational-immunology-modeling-the-immune-system) * [How to Set Hard Filters for Germline Variant Calling: A Practical Guide to GATK Best Practices](/knowledge/bioinformatics/how-to-set-hard-filters-for-germline-variant-calling-a-practical-guide-to-gatk-best-practices) ## References and Further Reading [1] [Signatures of mutational processes in human cancer.](https://pubmed.ncbi.nlm.nih.gov/23945592). Nature, 2013. [2] [Somatic mutation landscapes at single-molecule resolution.](https://pubmed.ncbi.nlm.nih.gov/33911282). Nature, 2021. [3] [Single-cell analysis of the somatic mutational landscape in human chondrocytes during aging and in osteoarthritis.](https://pubmed.ncbi.nlm.nih.gov/41249560). Nature aging, 2025. [4] [Accurate and sensitive mutational signature analysis with MuSiCal.](https://pubmed.ncbi.nlm.nih.gov/38361034). Nature genetics, 2024. [5] [Mutational signature analysis of chronic lymphocytic leukemia uncovering genomic patterns and prognostic implications.](https://pubmed.ncbi.nlm.nih.gov/40814292). American journal of clinical pathology, 2025. > This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.