How to Estimate False Discovery Rates in Proteomics: A Practical Guide to Target-Decoy and Percolator Approaches

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

False discovery rate (FDR) estimation is the statistical control mechanism that determines which peptide and protein identifications from a mass spectrometry experiment can be trusted for biological interpretation. This guide explains how to implement target-decoy competition (TDC) searches, calculate q-values, and use Percolator to improve FDR control in bottom-up proteomics workflows. The intended reader is a researcher, student, or laboratory professional who has generated tandem mass spectrometry data and needs to convert raw search results into a defensible list of identified peptides and proteins with known error rates.

Proteomics datasets have grown rapidly in size due to advances in separation techniques and mass analyzer performance, and this growth requires automated computational tools for high-throughput analysis and appropriate statistical control [<a href="#ref-1">1</a>]. Without proper FDR estimation, biological inferences drawn from proteomics data do not hold true and may be outright misleading [<a href="#ref-1">1</a>]. The practical problem is that database search algorithms assign matches to spectra, but those matches contain both true and false positives, and the overlap between the two groups can be considerable [<a href="#ref-1">1</a>]. FDR is the metric used for global confidence assessment of a large-scale proteomics dataset [<a href="#ref-1">1</a>].

This article covers the core principles of FDR estimation, the target-decoy workflow, the role of Percolator in rescoring, practical implementation steps, common failure patterns, and the limitations of each approach. The goal is to give you a workflow you can implement in your own data analysis pipeline with clear decision criteria at each step.

Key Takeaways

• False discovery rate (FDR) estimation in proteomics uses target-decoy competition (TDC), where spectra are searched against real target and shuffled or reversed decoy databases, and FDR is calculated as decoy hits divided by target hits at a score threshold.
• The standard TDC assumption that incorrect matches have equal probability of hitting target or decoy sequences holds approximately for most spectra, but only about 1.14 to 4.85 percent of spectra have an exact probability of 0.5, so FDR estimates are approximations, not exact values.
• Decoy databases must be constructed by reversal or shuffling of target sequences, searched with identical parameters as the target database, and appended to the target database as a single concatenated database to ensure unbiased scoring.
• Q-values, calculated as the minimum FDR at which a match would be accepted, allow you to apply a single threshold, typically 1% at the peptide level, to accept all matches with q-values below that threshold.
• Percolator, a machine learning tool, rescore peptide-spectrum matches by combining features like search engine scores, mass error, charge state, and peptide length into a single score, improving separation between true and false matches before FDR calculation.

At a Glance

The table below summarizes the main approaches to FDR estimation in proteomics, their primary use cases, and the key implementation considerations.

Why FDR Estimation Matters in Proteomics

Mass spectrometry-based proteomics generates spectra that must be matched to peptide sequences from a protein database. Each spectrum can match multiple candidate peptides with varying scores, and the search engine assigns the best match. The problem is that some of these matches are wrong. A spectrum might match a peptide sequence by chance, particularly when the database is large or the spectrum quality is poor.

FDR estimation addresses this problem by providing a statistical measure of how many identifications above a given score threshold are expected to be false. This is different from a per-match p-value because FDR considers the entire set of accepted identifications. When you report that your dataset has a 1% FDR at the peptide level, you are stating that approximately 1% of the accepted peptide-spectrum matches are expected to be incorrect.

The consequences of poor FDR control are substantial. If you set your threshold too leniently, your downstream biological conclusions may be based on proteins that are not actually present in your sample. If you set it too stringently, you lose true identifications and reduce the sensitivity of your experiment. The goal is to find the score threshold that gives you the maximum number of true identifications while keeping false positives at an acceptable level.

Statistical validation of false matches is necessary before making any biological interpretations from proteomics data [<a href="#ref-1">1</a>]. The biological inferences drawn from a dataset without proper FDR control do not hold true and may be misleading [<a href="#ref-1">1</a>]. This is why FDR estimation is not an optional quality check but a required component of any proteomics data analysis pipeline.

Core Principles of Target-Decoy Competition

Target-decoy competition is the most common technique for computing FDR estimates in tandem mass spectrometry proteomics [<a href="#ref-5">5</a>]. The approach works by searching observed spectra against a database of real target peptides and a database of shuffled or reversed decoy peptides [<a href="#ref-5">5</a>]. The decoy database is designed to have the same size and composition as the target database but with sequences that should not match any real peptides in the sample.

In the target-decoy search strategy, FDR is estimated by dividing the number of decoy hits at a given score threshold by the number of target hits, assuming equal search space for each database.

However, the assumption of equal probabilities is not always accurate. Research has shown that when spectra are actually identified incorrectly, most probabilities of the spectra matching a target or decoy peptide are not identical [<a href="#ref-2">2</a>]. In practice, most spectrum cases result in a probability of having the spectrum identified incorrectly as a target or decoy peptide close to 0.5, but only about 1.14 to 4.85 percent of the total spectra have an exact probability of 0.5 [<a href="#ref-2">2</a>]. This means that the standard TDC assumption holds approximately for most spectra but not exactly for any of them.

The practical implication is that TDC provides a useful approximation of the FDR, but the estimate may not be exact. The error is typically small, which is why TDC remains the standard approach. For most applications, the approximation is good enough. For applications requiring more precise FDR estimates, alternative approaches such as cTDS may be considered [<a href="#ref-2">2</a>].

Constructing Decoy Databases

The decoy database is the foundation of the target-decoy approach. The way you construct it affects the validity of your FDR estimates. The most common approaches are reversal and shuffling of target protein sequences.

Sequence Reversal

Sequence reversal involves reversing each protein sequence in the target database. This approach preserves the amino acid composition of each protein while destroying the biological sequence order. Reversal is straightforward to implement and is widely used because it produces decoy sequences that are clearly distinguishable from target sequences.

Sequence Shuffling

Sequence shuffling involves randomly permuting the amino acids in each protein sequence. This approach also preserves amino acid composition but can produce decoy sequences that are more similar to real peptides than reversed sequences. The choice between reversal and shuffling can affect the number of peptide-spectrum matches obtained at a given FDR threshold.

Research comparing different decoy database types found that the number of PSMs obtained with cTDS for reverse, pseudo-reverse, shuffle, and de Bruijn databases exceeded those obtained with TDS by about 0.001 to 0.132 percent, while the pseudo-shuffle database contained fewer PSMs compared to TDS by about 0.05 to 0.126 percent [<a href="#ref-2">2</a>]. These differences are small, but they illustrate that the choice of decoy construction method can have a measurable effect on your results.

Decoy Database Requirements

The decoy database must meet several requirements to produce valid FDR estimates. First, it must be searched with the same search parameters as the target database. Second, it must be approximately the same size as the target database. Third, the decoy sequences must not contain any real peptide sequences from the sample being analyzed.

The decoy database should be appended to the target database and searched as a single concatenated database. This ensures that each spectrum competes against both target and decoy sequences under identical search conditions. The search engine does not know which sequences are targets and which are decoys, so the scoring is unbiased.

Implementing Target-Decoy Searches

The implementation of a target-decoy search follows a standard workflow that can be adapted to most search engines and analysis pipelines.

Step 1: Prepare the Target Database

The target database should contain the protein sequences expected in your sample. For a human sample, this would be the human proteome. For a bacterial sample, this would be the proteome of the relevant bacterial species. The database should include common contaminants such as keratins and trypsin.

Step 2: Generate the Decoy Database

Generate decoy sequences from the target database using reversal or shuffling. Most search engines have built-in decoy generation options. If your search engine does not have this option, you can use external tools to generate decoy sequences.

Step 3: Concatenate Target and Decoy Databases

Combine the target and decoy sequences into a single database file. The concatenated database should be used for the search. Each protein entry should be labeled as either target or decoy so that you can separate the results after the search.

Step 4: Search the Spectra

Search your tandem mass spectra against the concatenated database using your chosen search engine. The search parameters should be identical to what you would use for a target-only search. The search engine will assign scores to each peptide-spectrum match, and the best match for each spectrum will be reported.

Step 5: Separate Target and Decoy Matches

After the search, separate the matches into target and decoy groups based on the protein labels. For each score threshold, count the number of target matches and decoy matches.

Step 6: Calculate FDR

For each score threshold, calculate the FDR as the number of decoy matches divided by the number of target matches. The q-value for each match is the minimum FDR at which that match would be accepted. You then select the score threshold that gives you your desired FDR, typically 1% at the peptide level and 1% at the protein level.

Step 7: Apply the Threshold

Apply the selected score threshold to your results. All target matches with scores above the threshold are accepted as identifications. All matches below the threshold are discarded.

Calculating Q-Values

The q-value is the minimum FDR at which a given match would be accepted. It is a per-match statistic that allows you to apply a consistent FDR threshold across your entire dataset. The q-value for a match is calculated by considering all matches with scores equal to or better than that match.

To calculate q-values from target-decoy search results, you first sort all matches by score in descending order. For each match, you calculate the FDR at that score threshold. The q-value for a match is the minimum FDR among all thresholds that include that match. This ensures that q-values are monotonic with score, meaning that a higher-scoring match always has a lower or equal q-value than a lower-scoring match.

The practical effect of q-value calculation is that you can set a single threshold, such as 1% FDR, and accept all matches with q-values below that threshold. This is simpler than manually selecting a score threshold and is the standard approach in proteomics data analysis.

Using Percolator for Rescoring

Percolator is a machine learning tool that improves the separation between true and false peptide-spectrum matches before FDR calculation. It works by combining multiple search features into a single score that better discriminates between correct and incorrect matches.

How Percolator Works

Percolator uses the target-decoy approach to train a classifier. The search results are labeled as target or decoy matches, and the classifier learns to distinguish between them based on features such as search engine scores, mass error, charge state, and peptide length. The trained classifier is then applied to all matches to produce a new score that is used for FDR calculation.

The key advantage of Percolator is that it can integrate information from multiple search engines and multiple features. A match that receives a moderate score from one search engine but a high score from another search engine may be more likely to be correct than a match that receives a moderate score from both. Percolator can capture this type of information and use it to improve the ranking of matches.

Integrating Percolator into Your Workflow

Percolator is typically run after the database search and before FDR calculation. The input is the search results file, and the output is a rescored results file with new scores and q-values. Most modern proteomics analysis platforms have built-in Percolator support, and it can also be run as a standalone tool.

The practical benefit of Percolator is that it allows you to identify more peptides and proteins at the same FDR threshold. By improving the separation between true and false matches, Percolator reduces the number of false matches above any given score threshold, which means you can lower the score threshold and accept more true matches while maintaining the same FDR.

Limitations of Percolator

Percolator performance depends on the quality of the features used for training and the composition of the training set. If the search results contain few decoy matches, the classifier may not have enough examples to learn effectively. If the features are not informative, the classifier may not improve over the original search scores.

Percolator also assumes that the target and decoy matches are correctly labeled. If the decoy database is poorly constructed or the search parameters are inappropriate, the training labels may be incorrect, and the classifier will learn the wrong patterns.

FDR Estimation at Different Levels

FDR can be estimated at multiple levels in a proteomics experiment: spectrum level, peptide level, and protein level. Each level requires a different approach to FDR calculation.

Spectrum-Level FDR

Spectrum-level FDR considers each peptide-spectrum match as an independent identification. This is the simplest level of FDR estimation and is often used as an intermediate step. The spectrum-level FDR is calculated by comparing the number of target and decoy spectrum matches above a score threshold.

Peptide-Level FDR

Peptide-level FDR considers each unique peptide sequence as an identification. Multiple spectra may match the same peptide, and the peptide is counted once regardless of how many spectra support it. Peptide-level FDR is typically lower than spectrum-level FDR because true peptides are often identified by multiple spectra, while false matches are more likely to be unique.

Protein-Level FDR

Protein-level FDR considers each protein as an identification. This is the most conservative level because a protein can be identified by multiple peptides, and the protein is counted once regardless of how many peptides support it. Protein-level FDR is typically higher than peptide-level FDR because proteins identified by a single peptide are more likely to be false.

The relationship between FDR levels is important to understand. Research in top-down proteomics found that an FDR calculated at the proteoform spectral match level under-reported the true FDR at the protein level by an average of 24-fold [<a href="#ref-3">3</a>]. This means that a 1% FDR at the spectral match level could correspond to a much higher FDR at the protein level. The same principle applies to bottom-up proteomics, where spectrum-level FDR does not directly translate to protein-level FDR.

Context-Dependent FDR

The context of the search affects FDR calculation. The context includes MS data quality, the database being interrogated, the search engine, and the parameters of the search [<a href="#ref-3">3</a>]. This means that FDR estimates are specific to the conditions under which they were calculated. If you change your search parameters, your FDR estimates may change even if your data is the same.

In top-down proteomics, there are four molecular levels of study: proteoform spectral match, protein, isoform, and proteoform [<a href="#ref-3">3</a>]. Each level requires its own FDR calculation, and the FDR at one level does not predict the FDR at another level [<a href="#ref-3">3</a>]. The TDCD_FDR_Calculator is an open-source tool that provides context-dependent FDR calculation for top-down proteomics and can be applied post-search to enhance the quality of results from any search engine [<a href="#ref-3">3</a>].

Alternative Approaches to FDR Estimation

While target-decoy competition is the standard approach, alternative methods exist for specific applications and for improving FDR accuracy.

Candidate Peptide Target-Decoy Strategy

The candidate peptide target-decoy strategy (cTDS) addresses the limitation of standard TDS by accounting for the probability that a spectrum is incorrectly identified as a target or decoy peptide [<a href="#ref-2">2</a>]. Standard TDS assumes these probabilities are identical, but research has shown they are not [<a href="#ref-2">2</a>]. cTDS calculates the probability for each spectrum and uses this information to produce more accurate FDR estimates.

Research using an entrapment sequence method demonstrated that for fixed FDR thresholds of 1 to 10 percent, the false match rate in cTDS is closer to the target FDR than the false match rate in TDS [<a href="#ref-2">2</a>]. This means that cTDS provides more accurate FDR control than standard TDS.

The practical benefit of cTDS is modest but measurable. In comparisons using the HEK293 dataset, the number of PSMs obtained with cTDS exceeded those obtained with TDS by about 0.001 to 0.132 percent for most decoy database types, with the pseudo-shuffle database containing fewer PSMs compared to TDS by about 0.05 to 0.126 percent [<a href="#ref-2">2</a>]. These differences are small, but they indicate that cTDS can identify slightly more true matches at the same FDR threshold.

Decoy-Free Approaches

Decoy-free approaches estimate FDR without using a decoy database. These approaches are particularly relevant for cross-linking mass spectrometry (XL-MS/MS), where the target-decoy approach has theoretical and practical limitations that impact estimation accuracy and increase run time [<a href="#ref-4">4</a>].

One decoy-free framework for XL-MS/MS relies on multi-sample mixtures of skew normal distributions, where the latent components correspond to the scores of correct peptide pairs, partially incorrect peptide pairs, and incorrect peptide pairs [<a href="#ref-4">4</a>]. The approach exploits the score distributions of first- and second-ranked peptide-spectrum matches for each experimental spectrum and estimates FDR using an expectation-maximization algorithm with constraints [<a href="#ref-4">4</a>]. Evaluation on ten datasets provided evidence that this decoy-free approach is theoretically sound and a viable alternative to the target-decoy approach [<a href="#ref-4">4</a>].

Decoy-free approaches are less established than TDC for standard bottom-up proteomics, but they represent an active area of research. For most applications, TDC remains the recommended approach because it is well understood, widely implemented, and validated across many studies.

Practical Workflow for FDR Estimation

The following workflow can be implemented in most proteomics data analysis pipelines. It assumes you have tandem mass spectrometry data and a protein database for your organism of interest.

Step 1: Prepare Your Data

Convert your raw mass spectrometry data to peak lists using your instrument software or a conversion tool. The peak lists should contain the mass-to-charge ratios and intensities of the fragment ions for each spectrum.

Step 2: Select Your Search Parameters

Choose your search engine and parameters. Common parameters include precursor mass tolerance, fragment mass tolerance, enzyme specificity, missed cleavages, and fixed and variable modifications. The parameters should be appropriate for your sample preparation and instrument.

Step 3: Generate the Concatenated Database

Generate the decoy database and concatenate it with the target database. Verify that the decoy database is the same size as the target database and that the sequences are properly labeled.

Step 4: Run the Database Search

Search your spectra against the concatenated database. The search engine will produce a list of peptide-spectrum matches with scores.

Step 5: Run Percolator

Run Percolator on the search results to rescore the matches. The output will include new scores and q-values.

Step 6: Calculate FDR at Each Level

Calculate FDR at the spectrum, peptide, and protein levels. Use the q-values from Percolator for spectrum and peptide level FDR. For protein level FDR, use a protein inference tool that accounts for the relationship between peptides and proteins.

Step 7: Apply Your FDR Threshold

Select your FDR threshold, typically 1% at the peptide and protein levels. Apply the threshold to your results and export the accepted identifications.

Step 8: Document Your Parameters

Record all search parameters, decoy construction methods, and FDR thresholds in your analysis documentation. This information is necessary for reproducibility and for interpreting your results.

Records and Measurements

Keeping detailed records of your FDR estimation process is essential for reproducibility and for troubleshooting problems. The following records should be maintained for each analysis.

Search Parameters

Record the search engine, version, and all search parameters. This includes precursor and fragment mass tolerances, enzyme specificity, missed cleavages, modifications, and any other settings that affect the search.

Decoy Construction

Record the method used to construct the decoy database, including whether sequences were reversed or shuffled and the tool used for decoy generation. If you used a custom decoy generation method, document the details.

Search Results Summary

Record the total number of spectra searched, the number of spectra matched to target peptides, the number matched to decoy peptides, and the number unmatched. These numbers provide context for interpreting your FDR estimates.

FDR Thresholds

Record the FDR thresholds applied at each level and the number of identifications accepted at each threshold. This information allows you to compare results across experiments and to assess the sensitivity of your analysis.

Percolator Settings

Record the Percolator version and settings, including the features used for training and any parameters that affect the rescoring process.

Common Failure Patterns

Several common mistakes can lead to incorrect FDR estimates. Recognizing these patterns can help you avoid them in your own analysis.

Improper Decoy Construction

The most common failure is improper decoy construction. If the decoy database is not the same size as the target database, or if the decoy sequences contain real peptide sequences, the FDR estimate will be incorrect. Always verify that the decoy database is properly constructed before running the search.

Searching Target and Decoy Separately

Some researchers search the target and decoy databases separately and then combine the results. This approach does not implement target-decoy competition because each spectrum is not competing against both target and decoy sequences. The FDR estimate from this approach is not valid. The target and decoy databases must be concatenated and searched together.

Using Different Search Parameters for Target and Decoy

The target and decoy databases must be searched with identical parameters. If different parameters are used, the scores are not comparable, and the FDR estimate is invalid.

Ignoring the Difference Between FDR Levels

Applying a spectrum-level FDR threshold to protein-level identifications is a common mistake. The FDR at the protein level is typically higher than the FDR at the spectrum level, so a 1% spectrum-level FDR does not guarantee a 1% protein-level FDR. Calculate FDR at each level separately and apply the appropriate threshold at each level.

Using an Inappropriate Decoy Database Size

The decoy database should be approximately the same size as the target database. If the decoy database is much smaller or larger, the FDR estimate will be biased. Most search engines automatically generate a decoy database of the correct size, but you should verify this if you are using a custom decoy generation method.

Failing to Account for Context

FDR estimates depend on the context of the search, including data quality, database composition, search engine, and search parameters [<a href="#ref-3">3</a>]. If you change any of these factors, your FDR estimates may change. Do not assume that FDR estimates from one experiment apply to another experiment with different conditions.

Limitations of FDR Estimation

FDR estimation is a statistical control mechanism, and it has inherent limitations that you should understand when interpreting your results.

The Assumption of Equal Target and Decoy Match Probabilities

Standard TDC assumes that when spectra are identified incorrectly, the probability of matching a target or decoy peptide is identical [<a href="#ref-2">2</a>]. Research has shown that this assumption is not exactly true [<a href="#ref-2">2</a>]. The error introduced by this assumption is usually small, but it can be larger in specific contexts.

The Dependence on Decoy Database Quality

The validity of TDC depends on the quality of the decoy database. If the decoy database is poorly constructed, the FDR estimate will be inaccurate. The decoy database must be similar in composition to the target database but must not contain any real peptide sequences.

The Difference Between FDR and False Match Rate

FDR is an estimate of the proportion of false matches among accepted identifications. The actual false match rate may differ from the estimated FDR. Research using entrapment sequence methods has shown that the false match rate in cTDS is closer to the target FDR than the false match rate in TDS, but neither method produces an exact false match rate [<a href="#ref-2">2</a>].

The Context Dependence of FDR

FDR calculation strongly depends on the context of the search, including MS data quality, the database being interrogated, the search engine, and the parameters of the search [<a href="#ref-3">3</a>]. This means that FDR estimates are not directly comparable across different experimental contexts.

The Level Dependence of FDR

FDR at one level does not predict FDR at another level. Research in top-down proteomics found that FDR calculated at the proteoform spectral match level under-reported the true FDR at the protein level by an average of 24-fold [<a href="#ref-3">3</a>]. Similar relationships exist in bottom-up proteomics, where spectrum-level FDR is typically lower than protein-level FDR.

Quality Controls and Validation

Several quality control measures can help you validate your FDR estimates and ensure that your identifications are reliable.

Entrapment Sequence Validation

Entrapment sequence validation involves adding known false sequences to the target database and checking whether they are identified at your FDR threshold. If entrapment sequences are identified at a rate higher than your FDR threshold, your FDR estimate may be too optimistic. This approach was used to demonstrate the accuracy of cTDS [<a href="#ref-2">2</a>].

Decoy Match Ratio

The ratio of decoy matches to target matches can indicate problems with your search. If the ratio is much higher than expected, your search parameters may be too lenient, or your decoy database may be problematic. If the ratio is much lower than expected, your search may be too stringent, or your decoy database may not be representative.

Mass Error Distribution

The mass error distribution of your accepted identifications should be centered near zero and have a narrow spread. If the distribution is shifted or broad, your mass calibration may be incorrect, and your identifications may be unreliable.

Peptide Length Distribution

The peptide length distribution of your accepted identifications should be consistent with your enzyme specificity and sample preparation. If the distribution is unusual, your search parameters may be incorrect.

Reproducibility Checks

If you have technical replicates, check that the same peptides and proteins are identified across replicates. Poor reproducibility may indicate problems with your FDR estimation or with your sample preparation.

Professional Escalation Criteria

Some situations require consultation with a bioinformatics specialist or a statistician. The following criteria indicate that you should seek professional assistance.

Persistent Discrepancies Between Expected and Observed FDR

If your decoy match ratio is consistently higher or lower than expected, or if your entrapment sequence validation shows a false match rate much higher than your FDR threshold, you should consult a specialist.

Unusual Score Distributions

If your score distributions for target and decoy matches overlap almost completely, or if the distributions have unusual shapes, your search parameters or decoy construction may be problematic.

Complex Experimental Designs

If your experiment involves multiple conditions, fractionation, or labeling strategies, the FDR estimation may require more sophisticated approaches than standard TDC. A specialist can help you design an appropriate analysis.

Top-Down or Cross-Linking Proteomics

Top-down and cross-linking proteomics have specific FDR estimation requirements that differ from standard bottom-up proteomics [<a href="#ref-3">3</a>][<a href="#ref-4">4</a>]. If you are working with these types of data, consult a specialist who has experience with these approaches.

Regulatory or Clinical Applications

If your proteomics data will be used for regulatory submissions or clinical decisions, the FDR estimation must meet higher standards of rigor. Consult a specialist to ensure that your analysis is appropriate for the intended use.

Training and Resources for FDR Estimation

Several official training resources can help you develop the skills needed for FDR estimation in proteomics.

The Galaxy Training Network provides accessible workflow training and analysis tutorials that cover proteomics data analysis, including database searching and FDR estimation [<a href="#ref-6">6</a>]. The tutorials are designed for researchers who want to learn practical analysis skills in a reproducible environment.

EMBL-EBI's freely accessible training courses cover database literacy, sequence analysis, structural biology, and expression data interpretation across the institute's core biological resources. These resources are useful for building a foundation in bioinformatics methods.

Software Carpentry and Data Carpentry workshops build core computational literacy - Unix shell, scripted data analysis, and Git version control - that underpins reproducible scientific computing. The command-line skills taught in these lessons are directly applicable to running database searches and FDR estimation tools.

Bioconductor's vignette-driven documentation model ensures every R package ships with a fully executable analysis workflow demonstrating best-practice reproducible genomics. Many proteomics analysis tools are available through Bioconductor, and the documentation includes examples of FDR estimation workflows.

nf-core enforces reproducibility through containerised pipeline releases, schema-validated parameter documentation, and automated CI testing prior to community publication. Several nf-core pipelines include proteomics analysis modules with built-in FDR estimation.

NCBI's publicly available technical documentation describes the scope, update cycles, and programmatic access patterns for its comprehensive suite of biological databases.

Frequently Asked Questions

What is the difference between FDR and a p-value?

A p-value is the probability of observing a result as extreme as the one observed, assuming the null hypothesis is true. It applies to a single test. FDR is the expected proportion of false positives among all accepted identifications. It applies to a set of tests. In proteomics, you are making thousands of identification decisions simultaneously, so FDR is the appropriate metric for global confidence assessment [<a href="#ref-1">1</a>].

How do I choose between sequence reversal and sequence shuffling for decoy construction?

Both reversal and shuffling are valid approaches for decoy construction. Research has shown small differences in the number of PSMs obtained with different decoy database types, with reverse, pseudo-reverse, shuffle, and de Bruijn databases producing slightly more PSMs than pseudo-shuffle databases at the same FDR threshold [<a href="#ref-2">2</a>]. The choice between reversal and shuffling is less important than ensuring that the decoy database is properly constructed and searched identically to the target database.

Why does my protein-level FDR differ from my peptide-level FDR?

Protein-level FDR is typically higher than peptide-level FDR because proteins identified by a single peptide are more likely to be false. The relationship between FDR levels is not straightforward. Research in top-down proteomics found that FDR calculated at the spectral match level under-reported the true FDR at the protein level by an average of 24-fold [<a href="#ref-3">3</a>]. You should calculate FDR at each level separately and apply the appropriate threshold at each level.

Can I use Percolator with any search engine?

Percolator is compatible with most popular database search engines. It takes the search results as input and produces rescored results as output. The features used for training are derived from the search results, so Percolator can work with any search engine that produces the required output format. Most modern proteomics analysis platforms have built-in Percolator support.

What is the difference between TDC and cTDS?

TDC assumes that when spectra are identified incorrectly, the probability of matching a target or decoy peptide is identical [<a href="#ref-2">2</a>]. cTDS relaxes this assumption by calculating the probability for each spectrum that it is incorrectly identified as a target or decoy peptide [<a href="#ref-2">2</a>]. Research has shown that cTDS produces false match rates closer to the target FDR than TDS [<a href="#ref-2">2</a>]. The practical difference in the number of PSMs obtained is small, typically less than 0.3 percent [<a href="#ref-2">2</a>].

When should I use a decoy-free approach instead of target-decoy?

Decoy-free approaches are primarily relevant for cross-linking mass spectrometry, where the target-decoy approach has theoretical and practical limitations [<a href="#ref-4">4</a>]. For standard bottom-up proteomics, target-decoy competition remains the recommended approach because it is well understood and widely validated. If you are working with XL-MS/MS data, you may consider a decoy-free approach that uses mixture models of score distributions [<a href="#ref-4">4</a>].

How do I know if my FDR estimate is accurate?

You can validate your FDR estimate using entrapment sequence methods, which involve adding known false sequences to the target database and checking whether they are identified at your FDR threshold [<a href="#ref-2">2</a>]. You can also check the decoy match ratio, mass error distribution, and peptide length distribution for signs of problems. If your validation checks indicate that your FDR estimate is inaccurate, you should review your search parameters and decoy construction.

What should I do if my decoy match ratio is unexpectedly high?

An unexpectedly high decoy match ratio may indicate that your search parameters are too lenient, your decoy database is problematic, or your data quality is poor. Review your search parameters and decoy construction. Check whether your mass calibration is correct. If the problem persists, consult a bioinformatics specialist.

Related Bioinformatics Guides

• Metagenomics Tools: A Practical Guide to Software and Pipelines
• Plasma Proteomics: From Sample Collection to Biomarker Discovery
• Pathway Enrichment Analysis for Proteomics: Tools and Interpretation
• Proteomics Analysis Tools: A Comparative Guide for Functional Interpretation
• Genomic Data Analysis Tools: A Comparative Guide for Researchers

References and Further Reading

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