# Evaluating Peptide Identification Confidence: How to Interpret Scores, E-Values, and q-Values from Search Engines Peptide identification in mass spectrometry-based proteomics relies on database search engines that assign statistical scores to peptide-spectrum matches (PSMs). These scores, including XCorr, Ion Score, and E-values, serve different purposes and cannot be compared directly across search engines. The practical problem researchers face is selecting thresholds that balance sensitivity and specificity without introducing excessive false discoveries. This article explains the statistical meaning of common scoring metrics, how post-processing tools such as Percolator refine these scores into q-values and posterior error probabilities, and how to set defensible cutoffs for your specific experimental context. ## The Problem with Raw Search Engine Scores Database search engines compare experimental tandem mass spectra against theoretical spectra generated from a protein sequence database. Each comparison produces a score that reflects how well the experimental spectrum matches a candidate peptide. Higher scores generally indicate better matches, but the raw score alone does not tell you the probability that the match is correct. Search engines use different scoring algorithms, which means an XCorr value from SEQUEST, an Ion Score from Mascot, and an E-value from MS-GF+ are not interchangeable. A score that indicates high confidence in one engine may be mediocre in another. This lack of standardization creates confusion when researchers attempt to apply universal thresholds across different platforms. The core issue is that raw scores are influenced by multiple factors unrelated to match quality, including spectrum quality, peptide length, charge state, and the size of the search space. A longer peptide will naturally accumulate more fragment ions and may receive a higher score simply because there are more opportunities for matches. Similarly, searching against a larger database increases the chance of random matches receiving high scores. ## Statistical Foundations of Peptide Scoring ### P-Values and Their Meaning A p-value in the context of peptide identification represents the probability of observing a score as extreme as the one obtained, assuming the match is random. Lower p-values indicate that the observed match is unlikely to occur by chance alone. However, p-values have an important limitation: they do not account for the multiple testing problem that arises when thousands of spectra are searched simultaneously. When you search 10,000 spectra against a database, even a p-value threshold of 0.01 would allow approximately 100 false positives by chance alone. This is why p-values alone are insufficient for large-scale proteomics experiments. ### E-Values and Expectation Scores E-values extend the concept of p-values by accounting for the number of comparisons performed. The E-value represents the expected number of random matches that would achieve a score equal to or better than the observed score across the entire search. An E-value of 0.01 means that you would expect 0.01 random matches with this score or better across all database searches performed. E-values are particularly useful because they scale with database size and search space. A peptide match with an E-value of 0.001 in a small database search carries different significance than the same E-value in a search against the entire human proteome. The E-value calculation incorporates the number of candidate peptides considered, making it a more interpretable metric than raw scores. ### Q-Values and False Discovery Rates Q-values address the multiple testing problem directly. The q-value for a peptide-spectrum match is the minimum false discovery rate (FDR) at which that match would be accepted as correct. In practical terms, if you set an FDR threshold of 1% and accept all PSMs with q-values below 0.01, you expect approximately 1% of your accepted identifications to be false positives. The false discovery rate framework is now standard in proteomics because it provides a principled way to control the proportion of errors among accepted identifications. Unlike p-values, which control the per-comparison error rate, q-values control the overall error rate across the entire set of accepted matches. ## How Search Engines Generate Scores ### SEQUEST and XCorr SEQUEST uses a cross-correlation approach to score peptide-spectrum matches. The XCorr value measures the correlation between the experimental spectrum and the theoretical spectrum of a candidate peptide. Higher XCorr values indicate stronger correlation and therefore higher confidence in the match. SEQUEST also calculates a deltaCn value, which represents the difference in correlation between the best match and the second-best match. A large deltaCn indicates that the top match is clearly better than alternatives, providing additional confidence. However, XCorr values are known to vary with peptide length and charge state, making fixed thresholds problematic across diverse datasets. ### Mascot and Ion Scores Mascot uses a probability-based scoring system where the Ion Score is calculated as -10 times the logarithm of the p-value. An Ion Score of 20 corresponds to a p-value of 0.01, while an Ion Score of 40 corresponds to a p-value of 0.0001. The identity threshold, typically displayed by Mascot, indicates the score required for significance at p < 0.05. The Ion Score incorporates information about the number of fragment ions matched, the peptide mass, and the database size. This makes it more interpretable than raw correlation scores, but the threshold for significance still depends on the specific search parameters and database used. ### MS-GF+ and Its Scoring Approach MS-GF+ uses a generative probabilistic model to score peptide-spectrum matches. The search engine calculates a spectral probability that represents the likelihood of observing the experimental spectrum given a particular peptide sequence. This probability is then converted into an E-value that accounts for the database search space. Research has shown that MS-GF+ provides efficient scoring but offers limited statistical estimates about peptide-spectrum matches on its own. The original publication describing MS-GF+ processing with Percolator demonstrated that post-processing with Percolator increased the number of identified peptides across a wide variety of datasets. This finding highlights the importance of post-processing tools for improving both the number and reliability of identifications. ## The Role of Post-Processing Tools ### Percolator and Semi-Supervised Learning Percolator uses a semi-supervised machine learning approach to distinguish correct from incorrect peptide-spectrum matches. The algorithm learns a discriminant function that combines multiple features from the search engine output, including raw scores, delta scores, and peptide properties, to separate true matches from decoy matches. The key advantage of Percolator is that it directly reports p-values, q-values, and posterior error probabilities for peptide-spectrum matches, peptides, and proteins. These statistical estimates are useful for the entire proteomics community because they provide a consistent framework for confidence assessment across different search engines. Percolator works by using target-decoy competition. The search is performed against a database containing both target protein sequences and decoy sequences, typically reversed or shuffled versions of the target sequences. Matches to decoy sequences are known to be false, providing a training set for the machine learning algorithm. The algorithm then learns to distinguish target matches from decoy matches based on the available features. ### Target-Decoy Database Searching The target-decoy approach is fundamental to modern false discovery rate estimation. By searching against a database that contains known false sequences, researchers can estimate the number of false positives among accepted target matches. The assumption is that random matches to decoy sequences occur at the same rate as random matches to target sequences. This approach requires careful database construction. Decoy sequences should be similar in composition to target sequences but clearly distinguishable as false. Common approaches include reversing protein sequences or shuffling amino acids while preserving composition. The target-decoy strategy provides an empirical estimate of the false discovery rate that does not rely on theoretical models of score distributions. ### Linear Discriminant Analysis for Score Combination Some pipelines use linear discriminant analysis to combine multiple peptide scores into a single discriminant score with maximum discriminating power. This approach was demonstrated for phosphopeptide identifications, where an expectation maximization algorithm estimated the joint distribution of peptide scores, and linear discriminant analysis determined how to optimally combine SEQUEST scores into a single discriminant score. The resulting discriminant score was used to calculate p-values and q-values for each phosphopeptide identification, providing a robust estimate of the false discovery rate. This approach is particularly valuable when search engines produce multiple scores that capture different aspects of match quality, because combining them can improve discrimination between correct and incorrect matches. ## At a Glance: Score Interpretation Reference | Metric | What It Measures | Interpretation Guidance | Practical Threshold Considerations | |--------|-----------------|------------------------|----------------------------------| | XCorr (SEQUEST) | Cross-correlation between experimental and theoretical spectra | Higher values indicate better correlation, values vary with peptide length and charge state | Avoid fixed thresholds across diverse datasets, use charge-state-specific cutoffs or post-processing | | Ion Score (Mascot) | -10 times log of p-value | Score of 20 equals p = 0.01, identity threshold indicates significance at p < 0.05 | Threshold depends on database size and search parameters, use the displayed identity threshold as a starting point | | E-value (MS-GF+) | Expected number of random matches with equal or better score | Lower values indicate fewer expected random matches, scales with database size | E-values are comparable across searches with the same database, use post-processing for FDR control | | q-value (Percolator) | Minimum FDR at which a match is accepted | Directly controls the proportion of false positives among accepted matches | Set q-value threshold based on your acceptable FDR, 1% is common for discovery proteomics | | Posterior error probability (Percolator) | Probability that a match is incorrect | Lower values indicate higher confidence | Use in combination with q-values for filtering, PEP can be used for protein-level scoring | ## Setting Score Thresholds for Your Experiment ### Define Your Acceptable False Discovery Rate The first decision in setting score thresholds is determining your acceptable false discovery rate. This choice depends on your experimental goals and the consequences of false identifications. Discovery experiments aimed at generating hypotheses may tolerate a 5% FDR, while validation experiments or clinical studies may require a 1% or even 0.1% FDR. The false discovery rate framework provides a direct way to control the proportion of errors among your accepted identifications. When you set an FDR threshold of 1%, you are stating that you accept approximately 1% false positives among your reported identifications. This is a more meaningful guarantee than a p-value threshold because it directly addresses the multiple testing problem. ### Use Post-Processing for Consistent Confidence Assessment Raw search engine scores are difficult to interpret consistently across datasets and search engines. Post-processing tools such as Percolator provide a consistent framework for confidence assessment by converting raw scores into q-values and posterior error probabilities. These tools also increase the number of identified peptides by rescoring matches based on learned discrimination functions. The research demonstrating MS-GF+ processing with Percolator showed increased peptide identifications across a wide variety of datasets. This improvement occurs because Percolator learns to recognize features that distinguish correct from incorrect matches, allowing it to rescue borderline matches that would be rejected by simple score thresholds. ### Consider the Size of Your Search Space The number of candidate peptides in your database search directly affects the significance of any given score. A larger search space increases the chance of random matches achieving high scores. E-values account for this by scaling with the number of comparisons performed, but raw scores such as XCorr do not. When comparing results across experiments, ensure that the search space is comparable. Searching against a species-specific database versus the entire NCBI non-redundant database will produce different score distributions. The NCBI provides access to a wide range of sequence databases and search systems that can be used for proteomics analysis, and the choice of database should be documented in your methods. ### Validate Thresholds with Decoy Matches The target-decoy approach provides an empirical validation of your score thresholds. After setting a threshold, examine the number of decoy matches that pass the threshold. The ratio of decoy to target matches passing the threshold provides an estimate of the false discovery rate. If your observed FDR exceeds your acceptable threshold, you need to increase your score cutoff. This validation step is essential because theoretical models of score distributions may not perfectly match your experimental data. Spectrum quality, sample complexity, and instrument settings all influence the actual distribution of scores. Empirical validation using decoy matches provides a direct measurement of the false discovery rate for your specific dataset. ## Practical Workflow for Confidence Assessment ### Step 1: Perform the Database Search Run your database search using your chosen search engine. Document the search parameters, including the database version, precursor mass tolerance, fragment mass tolerance, enzyme specificity, and allowed modifications. These parameters affect the search space and therefore the significance of scores. Ensure that your database contains both target and decoy sequences. The target-decoy approach is essential for false discovery rate estimation. Most modern search engines and analysis platforms support automatic decoy generation, but you should verify that decoys are properly constructed and that the search results include decoy matches. ### Step 2: Run Post-Processing Feed your search results into a post-processing tool such as Percolator. The post-processor will rescore your peptide-spectrum matches and calculate q-values and posterior error probabilities. This step is critical for consistent confidence assessment across different search engines and datasets. The post-processing step also increases the number of identified peptides by rescoring matches based on learned discrimination functions. This improvement is particularly valuable for challenging datasets with many borderline matches. ### Step 3: Examine Score Distributions Before setting thresholds, examine the distribution of scores for target and decoy matches. Plot histograms of raw scores, q-values, and posterior error probabilities for both target and decoy matches. This visualization helps you understand the separation between correct and incorrect matches in your dataset. A good separation between target and decoy score distributions indicates that your search engine and post-processing pipeline are working effectively. Poor separation suggests that your search parameters may need adjustment or that your data quality is insufficient for confident identification. ### Step 4: Set Thresholds Based on FDR Set your q-value threshold based on your acceptable false discovery rate. For discovery experiments, a 1% FDR is a common starting point. For more stringent applications, use a 0.1% FDR. The q-value threshold directly controls the proportion of false positives among your accepted identifications. After setting the threshold, check the number of decoy matches that pass the threshold. The observed FDR should be close to your target. If the observed FDR is much higher than expected, your search parameters or post-processing may need adjustment. ### Step 5: Validate with Known Standards If your experiment includes known standards, such as a protein mixture with known composition, validate your thresholds using these standards. Check that the expected proteins are identified and that the false discovery rate among known standards is acceptable. This validation step provides additional confidence that your thresholds are appropriate for your experimental context. Known standards can also help you calibrate thresholds when working with a new instrument, new search engine, or new sample type. ## Records and Measurements for Confidence Assessment ### Documenting Search Parameters Maintain detailed records of all search parameters for each analysis. This documentation should include the search engine version, database version and source, decoy construction method, precursor and fragment mass tolerances, enzyme specificity, allowed modifications, and any other parameters that affect the search space. The NCBI provides access to sequence databases and search systems that are commonly used in proteomics analysis. Documenting the specific database version is important because database updates can affect search results and score distributions. ### Tracking Score Distributions Across Batches Record the distribution of scores for each batch of samples analyzed. This tracking helps you identify batch effects or instrument drift that may affect identification confidence. If score distributions shift significantly between batches, your thresholds may need adjustment. Create a standard report format that includes the number of spectra searched, the number of peptide-spectrum matches at various score thresholds, the number of decoy matches, and the estimated false discovery rate. This report format enables consistent comparison across experiments and batches. ### Recording Post-Processing Output Document the post-processing output, including the number of peptides and proteins identified at your chosen FDR threshold. Record the q-value distributions and posterior error probabilities for your accepted identifications. This information is essential for reporting your results in publications and for comparing results across experiments. The post-processing output also provides information about the quality of your identifications beyond simple counts. Examining the distribution of posterior error probabilities among accepted identifications can reveal whether your threshold is appropriately stringent or whether you are accepting many borderline matches. ## Common Failure Patterns in Score Interpretation ### Applying Universal Thresholds Across Search Engines A common mistake is applying a score threshold learned from one search engine to results from another engine. XCorr thresholds from SEQUEST cannot be applied to Ion Scores from Mascot or E-values from MS-GF+. Each search engine uses a different scoring algorithm, and the score distributions are not directly comparable. The solution is to use post-processing tools that convert raw scores into a consistent statistical framework. Q-values and posterior error probabilities from Percolator are comparable across search engines because they are based on the target-decoy approach instead of the raw score distribution. ### Ignoring Database Size Effects The significance of a given score depends on the size of the search space. A score that is highly significant in a search against a small database may be less significant in a search against a large database. E-values account for this effect, but raw scores do not. When comparing results across experiments, ensure that the search space is comparable. If you change databases or search parameters, re-evaluate your thresholds. The target-decoy approach provides an empirical way to assess the false discovery rate for your specific search space. ### Using P-Values Without Multiple Testing Correction P-values alone do not account for the multiple testing problem inherent in large-scale proteomics. Searching thousands of spectra against a large database creates many opportunities for random matches. A p-value threshold of 0.01 would allow many false positives by chance alone. The solution is to use q-values or posterior error probabilities that account for multiple testing. The false discovery rate framework provides a principled way to control the proportion of errors among accepted identifications. ### Overlooking Post-Processing Benefits Some researchers skip post-processing and rely on raw search engine scores for confidence assessment. This approach is problematic because raw scores are difficult to interpret consistently and do not provide direct false discovery rate estimates. Post-processing tools such as Percolator provide several benefits. They increase the number of identified peptides by rescoring matches based on learned discrimination functions. They also provide consistent statistical estimates, including p-values, q-values, and posterior error probabilities, for peptide-spectrum matches, peptides, and proteins. ### Confusing Posterior Error Probability with Q-Value Posterior error probability and q-value are related but distinct concepts. The posterior error probability for a peptide-spectrum match is the probability that the match is incorrect. The q-value is the minimum false discovery rate at which the match would be accepted. A match with a low posterior error probability may still have a relatively high q-value if many other matches with similar scores are also accepted. The q-value accounts for the cumulative effect of accepting multiple matches, while the posterior error probability applies to individual matches. ## Limitations of Score-Based Confidence Assessment ### Incomplete Databases and Novel Peptides Score-based confidence assessment assumes that the correct peptide is present in the database being searched. If the database is incomplete or does not contain the correct peptide sequence, even a perfect spectrum will not produce a high-scoring match. This limitation is particularly relevant for cancer proteomics, where non-canonical peptides arising from genomic dysregulation may be absent from reference protein sequence databases. Research on spectral clustering approaches has shown that many acquired spectra in any given experiment are not identified, with causes including poor-quality spectra, amino acid modifications, nonsynonymous variants, and non-canonical peptides absent from the reference database. When working with samples where novel peptides are expected, consider using spectral clustering or other approaches to identify peptides missed by standard database searching. These approaches can improve the identification of novel, sample-specific peptides that are absent from reference databases. ### Spectrum Quality and Fragmentation The quality of the experimental spectrum directly affects the confidence of peptide identifications. Poor-quality spectra with limited fragment ion coverage produce lower scores and higher uncertainty. Factors such as incomplete fragmentation, co-eluting peptides, and chemical noise can degrade spectrum quality. Score thresholds cannot compensate for poor spectrum quality. If your dataset contains many low-quality spectra, you may need to apply more stringent thresholds to maintain an acceptable false discovery rate. Alternatively, you may need to optimize your sample preparation and chromatography to improve spectrum quality. ### Post-Translational Modifications Post-translational modifications complicate peptide identification because they change the mass of the peptide and affect fragmentation patterns. Search engines must consider the possibility of modifications, which increases the search space and affects score distributions. For phosphopeptide identifications, the ambiguity of phosphate localization adds another layer of complexity. Research on phosphopeptide identification has demonstrated the importance of reanalyzing identifications with ambiguous phosphate assignments to determine the most likely arrangement of the phosphate groups. This reanalysis improves the accuracy of the final identifications. ### Search Space and Computational Cost Increasing the search space to accommodate modifications, variants, or non-canonical peptides increases computational cost and affects score significance. A larger search space means more opportunities for random matches, which requires more stringent thresholds to maintain the same false discovery rate. The tradeoff between search space and sensitivity is a fundamental challenge in proteomics. Including more potential modifications increases the chance of identifying modified peptides but also increases the chance of false positives. Post-processing tools can help manage this tradeoff by learning to distinguish correct from incorrect matches based on multiple features. ## Quality Controls for Confidence Assessment ### Replicate Analysis Analyze replicate samples to assess the reproducibility of your identifications. Peptides identified consistently across replicates are more likely to be correct than those identified in only one replicate. The false discovery rate framework does not directly account for reproducibility, so replicate analysis provides complementary information. When reporting results, distinguish between peptides identified in all replicates, peptides identified in some replicates, and peptides identified in only one replicate. This distinction provides useful information about the confidence of your identifications. ### Known Standard Validation Include known standards in your experimental design to validate your identification pipeline. Standards with known protein composition provide a ground truth for assessing the accuracy of your identifications. If your pipeline fails to identify expected proteins or identifies unexpected proteins, your thresholds or search parameters may need adjustment. The use of universal protein standards in a defined background, as described in spectral clustering research, provides a controlled system for evaluating identification performance. This approach allows you to assess sensitivity and specificity under controlled conditions. ### Cross-Platform Validation If possible, validate key identifications using an orthogonal approach. For example, confirm peptide identifications using synthetic peptides or alternative fragmentation methods. Cross-platform validation provides additional confidence beyond score-based assessment. This validation is particularly important for novel or unexpected identifications that may have clinical significance. The research on non-canonical peptides in cancer highlights the importance of validating novel identifications before drawing biological conclusions. ### Monitoring False Discovery Rate Over Time Track the false discovery rate across experiments and batches to identify trends or anomalies. A sudden increase in the false discovery rate may indicate instrument problems, sample contamination, or database issues. Regular monitoring helps maintain consistent data quality. Create a quality control report for each batch that includes the number of spectra searched, the number of identifications at your chosen FDR threshold, the observed false discovery rate, and any anomalies or deviations from expected patterns. ## Safety and Regulatory Context for Confidence Assessment ### Data Integrity and Reproducibility Score-based confidence assessment is part of the broader framework of data integrity and reproducibility in proteomics. The field has established standards for reporting identifications, including the false discovery rate and the search parameters used. Adhering to these standards is essential for reproducible research. Training resources from organizations such as the Galaxy Training Network and The Carpentries provide foundational instruction in reproducible analysis practices. These resources emphasize the importance of documenting analysis steps, using version control, and maintaining reproducible workflows. ### Reporting Standards for Publications When reporting peptide identifications in publications, include the search engine version, database version, search parameters, and the false discovery rate threshold used. This information allows readers to assess the confidence of your identifications and to reproduce your analysis. The false discovery rate should be reported for both peptide and protein levels. Protein-level false discovery rates are typically more stringent than peptide-level rates because protein inference adds another layer of complexity. ### Workflow Reproducibility Use workflow management systems to ensure that your analysis pipeline is reproducible. Community standards such as those promoted by the nf-core project provide guidelines for building reproducible bioinformatics workflows. These standards emphasize version control, containerization, and documentation. Reproducible workflows are essential for confidence assessment because they ensure that the same inputs produce the same outputs. If your analysis pipeline is not reproducible, your confidence assessment cannot be independently verified. ### Training and Competency Ensure that researchers performing proteomics analysis have adequate training in statistical concepts and bioinformatics tools. The EMBL-EBI provides training resources for bioinformatics analysis, including practical education in data analysis. Foundational training in computing and data skills is available through The Carpentries lessons. Inadequate training in statistical concepts can lead to misinterpretation of scores and inappropriate threshold selection. Investing in training improves the quality of your confidence assessment and reduces the risk of reporting false identifications. ## Professional Escalation Criteria ### When to Consult a Bioinformatics Specialist If you are uncertain about the appropriate score thresholds for your experiment, consult a bioinformatics specialist or a core facility scientist. These experts can help you select appropriate thresholds based on your experimental design and data quality. Signs that you should escalate include poor separation between target and decoy score distributions, unexpected false discovery rates, or difficulty reproducing identifications across replicates. These issues may indicate problems with your search parameters, database, or data quality that require expert attention. ### When to Re-Evaluate Your Search Parameters If your observed false discovery rate is much higher than expected, re-evaluate your search parameters. Common issues include incorrect precursor mass tolerance, inappropriate enzyme specificity, or missing modifications. Adjusting these parameters can improve the separation between correct and incorrect matches. The target-decoy approach provides a direct measurement of the false discovery rate for your specific search parameters. If the observed false discovery rate exceeds your acceptable threshold, you need to adjust your parameters or thresholds. ### When to Question Your Database Choice If you are working with a sample type that may contain novel peptides, question whether your database is appropriate. Reference databases may not contain non-canonical peptides or sample-specific variants. Consider using spectral clustering or other approaches to identify peptides missed by standard database searching. The research on spectral clustering for cancer proteomics demonstrated that consensus spectra from spectral clusters can lead to better peptide-spectrum matches in terms of coverage and scoring confidence. This approach can identify non-canonical peptides missed by standard database searching. ### When to Seek Additional Validation If you are reporting novel or unexpected identifications, seek additional validation before drawing biological conclusions. This validation may include synthetic peptide confirmation, orthogonal fragmentation methods, or targeted mass spectrometry approaches. The research on umami peptide identification in non-alcoholic beer demonstrated a multi-step validation approach that combined machine learning prediction, molecular docking, and sensory validation. This approach provides a model for validating peptide identifications beyond score-based confidence assessment. ## Frequently Asked Questions ### What is the difference between a p-value and a q-value in peptide identification? A p-value represents the probability of observing a score as extreme as the one obtained, assuming the match is random. A q-value represents the minimum false discovery rate at which a match would be accepted as correct. The key difference is that q-values account for the multiple testing problem inherent in large-scale proteomics. When you search thousands of spectra, many p-values will be small by chance alone. Q-values control the proportion of false positives among accepted identifications, providing a more meaningful guarantee for large datasets. ### Why do different search engines produce different scores for the same spectrum? Different search engines use different scoring algorithms that emphasize different aspects of the peptide-spectrum match. SEQUEST uses cross-correlation, Mascot uses probability-based scoring, and MS-GF+ uses a generative probabilistic model. These algorithms calculate scores based on different features and assumptions, so the scores are not directly comparable. This is why post-processing tools such as Percolator are valuable: they convert raw scores from different engines into a consistent statistical framework based on q-values and posterior error probabilities. ### How do I choose an appropriate false discovery rate threshold? The choice of false discovery rate threshold depends on your experimental goals and the consequences of false identifications. Discovery experiments aimed at generating hypotheses may tolerate a 5% FDR, while validation experiments or clinical studies may require a 1% or 0.1% FDR. Consider the downstream consequences of false identifications. If you are making biological claims based on identified peptides, use a more stringent threshold. If you are generating candidate lists for further validation, a less stringent threshold may be acceptable. ### What is the target-decoy approach and why is it important? The target-decoy approach involves searching your spectra against a database containing both target protein sequences and decoy sequences, typically reversed or shuffled versions of the target sequences. Matches to decoy sequences are known to be false, providing a training set for machine learning algorithms and an empirical estimate of the false discovery rate. This approach is important because it provides a direct measurement of the false discovery rate for your specific dataset, instead of relying on theoretical models of score distributions. ### Can I compare E-values across different database searches? E-values are designed to be comparable across searches because they account for the number of comparisons performed. An E-value of 0.01 means that you would expect 0.01 random matches with this score or better across the entire search. However, E-values are only comparable when the search space is similar. Searching against a larger database increases the number of candidate peptides and therefore affects E-values. When comparing results across experiments, ensure that the database and search parameters are comparable. ### What is the role of Percolator in peptide identification? Percolator is a post-processing tool that uses semi-supervised machine learning to distinguish correct from incorrect peptide-spectrum matches. It learns a discriminant function that combines multiple features from the search engine output to separate true matches from decoy matches. Percolator directly reports p-values, q-values, and posterior error probabilities for peptide-spectrum matches, peptides, and proteins. Research has shown that Percolator processing increases the number of identified peptides across a wide variety of datasets. ### How do I handle ambiguous post-translational modification localization? Ambiguous modification localization is a common challenge, particularly for phosphopeptide identifications. Research has demonstrated the importance of reanalyzing identifications with ambiguous phosphate assignments to determine the most likely arrangement of the phosphate groups. This reanalysis improves the accuracy of the final identifications. Consider using specialized tools or pipelines that address modification localization ambiguity when working with modified peptides. ### What should I do if my observed false discovery rate is higher than expected? If your observed false discovery rate is higher than expected, first check your search parameters. Common issues include incorrect precursor mass tolerance, inappropriate enzyme specificity, or missing modifications. Adjust these parameters and re-run the search. If the problem persists, examine your spectrum quality and consider whether your sample preparation or chromatography needs optimization. If you are still unable to achieve an acceptable false discovery rate, consult a bioinformatics specialist for assistance. ## Related Bioinformatics Guides - [How to Interpret Gene Set Enrichment Analysis Results](/knowledge/bioinformatics/how-to-interpret-gene-set-enrichment-analysis-results) - [Volcano Plot Proteomics: How to Create and Interpret Them Effectively](/knowledge/bioinformatics/volcano-plot-proteomics-how-to-create-and-interpret-them-effectively) - [Single-Cell Annotation: A Workflow for Cell Type Identification](/knowledge/bioinformatics/single-cell-annotation-a-workflow-for-cell-type-identification) - [Genomic Data Analysis Tools: A Comparative Guide for Researchers](/knowledge/bioinformatics/genomic-data-analysis-tools-a-comparative-guide-for-researchers) - [Evaluating Genome Assembly Quality: Metrics and Tools](/knowledge/bioinformatics/evaluating-genome-assembly-quality-metrics-and-tools) ## 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. - [Fast and accurate database searches with MS-GF+Percolator.](https://pubmed.ncbi.nlm.nih.gov/24344789). Journal of proteome research, 2014. - [Linear discriminant analysis-based estimation of the false discovery rate for phosphopeptide identifications.](https://pubmed.ncbi.nlm.nih.gov/18422353). Journal of proteome research, 2008. - [Building Consensus TMT Mass Spectra from Spectral Clusters to Explore the Unknown Cancer Proteome](https://doi.org/10.21203/rs.3.rs-9601247/v1). 2026. - [Full-DIA enables complete single-cell proteomics from diaPASEF using deep learning.](https://doi.org/10.1186/s13059-026-04087-x). 2026. - [Mining and Validation of Novel Umami Peptides in Non-Alcoholic Beer by Integrating Machine Learning Prediction, Molecular Docking, and Sensory Validation, and Their Multidimensional Sensory Impacts on Beer Body.](https://doi.org/10.3390/foods15101671). 2026. > This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.