Evaluating Proteogenomic Search Results: How to Assess Confidence and Avoid False Positives

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

Proteogenomic searches integrate mass spectrometry proteomics data with genomic and transcriptomic sequence databases to identify variant peptides, including single amino acid variants (SAAVs) and products of single nucleotide polymorphisms (SNPs). The central problem addressed in this article is that these searches routinely produce false-positive identifications because the customized sequence databases used are large, complex, and contain many peptide sequences that differ from the reference proteome by only one amino acid. Researchers need a practical framework for evaluating peptide-spectrum matches (PSMs) that includes false discovery rate (FDR) estimation, manual spectrum inspection, and orthogonal validation using synthetic peptides or targeted mass spectrometry. This article provides that framework, with concrete decisions, records, and escalation criteria for biology students, researchers, and laboratory professionals.

Key Takeaways

• Proteogenomic searches routinely produce false-positive variant peptide identifications because customized sequence databases are large and contain peptides differing from the reference proteome by only one amino acid.
• In a reanalysis of 2.8 million spectra, only 23.8% of 12,705 variant peptide-spectrum matches were classified as confident, while 65.1% were doubtful after rigorous quality evaluation.
• Researchers should include an entrapment database containing sequences known to be absent from the sample in every proteogenomic search to measure and control false discovery rates.
• Class-specific false discovery rate estimation must be computed separately for variant peptides, since the FDR among variants can be 20% or higher even when the global FDR is 1%.
• Manual spectrum inspection should verify fragment ion series coverage, variant-supporting ions, absence of unexplained dominant peaks, and mass error, with all decisions recorded in a written standard operating procedure.

The Scale of the False Positive Problem in Proteogenomics

Proteogenomics combines high-throughput nucleic acid analysis with protein mass spectrometry to identify protein-level evidence of genomic variation. A primary goal is the reliable identification of single amino acid substitutions, which represent a main type of coding genome variant. Exact knowledge of deviations from the consensus genome has applications in studying mutated protein expression in cancer, deciphering heterozygosity mechanisms, identifying neoantigens for anticancer vaccine production, and searching for RNA editing sites at the proteome level [<a href="#ref-1">1</a>].

The difficulty is that extracting information on single-point protein variation from high-resolution mass spectrometry data is technically challenging. A significant problem in proteogenomic analysis is the presence of high levels of false-positive results for variant-containing peptides [<a href="#ref-1">1</a>]. When a search database contains both the reference (wild-type) peptide sequence and one or more variant sequences that differ by a single amino acid, the search engine must distinguish between a true variant and a mis-assignment caused by noise, co-eluting peptides, or post-translational modifications that produce similar mass shifts.

The scale of this problem is documented in large-scale reanalysis studies. In one reanalysis of approximately 2.8 million spectra from two public proteomics datasets, a total of 12,705 variant PSMs were identified. When those PSMs were evaluated with a dedicated quality-control tool, only 3,028 (23.8%) were classified as confident, 1,409 (11.1%) as semi-confident, and 8,268 (65.1%) as doubtful [<a href="#ref-2">2</a>]. This means that nearly two-thirds of the variant PSMs initially identified by standard search pipelines did not survive rigorous quality evaluation. Researchers who skip this evaluation step risk reporting results that are predominantly false positives.

Database Construction and Its Effect on False Discovery

The design of the sequence database used for searching is the first decision point that determines whether false positives will be controlled. Proteogenomic databases are typically constructed by adding variant peptide sequences to a reference proteome database. The source of those variants matters.

Whole Exome Sequencing Databases

A common approach is to build a mutation database from whole exome sequencing (WES) data generated from the same sample being analyzed by mass spectrometry. This approach has limitations. WES-based databases can miss potential neoantigens due to limitations in sequencing depth and sample quality [<a href="#ref-3">3</a>]. In a study of HLA-I immunopeptidome identification from HepG2 cells, a WES-based database identified only 1 mutant peptide, while a database constructed from the COSMIC mutation database identified 5 immunopeptides [<a href="#ref-3">3</a>]. The WES-based database excluded certain mutant peptides because of sequencing depth limitations or sample heterogeneity [<a href="#ref-3">3</a>].

For researchers deciding between sample-specific WES databases and public mutation databases, the tradeoff is between specificity and completeness. A sample-specific database is more biologically relevant to the tissue being studied, but it is only as complete as the sequencing coverage allows. A public database such as COSMIC provides broader coverage of known mutations but includes variants that may not be present in the sample, which increases database size and can increase the search space.

Customized Sequence Databases for Specific Cell Lines

Some cell lines have well-characterized genomes that differ substantially from the human reference genome. HeLa cells are a notable example. Researchers studying SAAVs in HeLa cells have developed customized sequence databases that incorporate SAAV sequences from the HeLa genome [<a href="#ref-4">4</a>]. This approach ensures that the database contains the actual variant sequences present in the sample, instead of relying on the reference genome, which may not match the cell line.

The key decision for database construction is whether to include all possible variants from a genomic source or to filter variants by quality and predicted effect. Including every SNP from a database without filtering will produce a very large search space, which increases the computational burden and can increase false discoveries. Filtering by variant quality, allele frequency, and predicted protein-coding effect reduces the database to variants that are plausibly expressed and detectable.

Entrapment Databases as a Control

Entrapment databases are a practical control for measuring false discovery in proteogenomic searches. An entrapment database contains sequences that are known to be absent from the sample, such as proteins from a distantly related organism. When a search returns a high number of identifications from the entrapment sequences, this indicates that the search parameters are too permissive and that false positives are likely.

In the assessment of DIA-MS for SAAV identification, entrapment databases were used to evaluate the false discovery match ratio (FDMR) across different search engine pipelines [<a href="#ref-4">4</a>]. The study found that DIA-MS searches maintained consistent identification of SAAV peptides despite varying sizes of the entrapment database [<a href="#ref-4">4</a>]. This consistency is a desirable property because it indicates that the search results are not simply a function of database size. However, the same study found that different search engines performed differently, with Fragpipe-MSFragger demonstrating the most conservative and effective performance, exhibiting the lowest false discovery match ratio [<a href="#ref-4">4</a>].

For routine practice, researchers should include an entrapment database in every proteogenomic search. The entrapment sequences should be clearly labeled so that they can be filtered out of the final results. The number of entrapment identifications should be recorded and reported. If the entrapment identification rate rises above an acceptable threshold, the search parameters should be made more stringent before any variant identifications are accepted.

False Discovery Rate Estimation in Proteogenomic Searches

FDR estimation is the statistical backbone of proteogenomic confidence assessment. The standard target-decoy approach, in which searches are performed against a combined database of real (target) sequences and reversed or shuffled (decoy) sequences, is widely used. However, proteogenomic searches present specific challenges for FDR estimation because variant peptides are often present in the database at very low frequencies, and the decoy database may not accurately model the distribution of variant peptides.

Class-Specific FDR

Standard global FDR control treats all peptide identifications equally. In proteogenomics, this is insufficient because variant peptides are much rarer than reference peptides, and the false discovery rate among variant peptides can be much higher than the global FDR. Class-specific FDR, in which the FDR is estimated separately for variant peptides and reference peptides, is a necessary refinement [<a href="#ref-2">2</a>].

The practical implication is that a search may report a global FDR of 1%, but the FDR among variant peptides specifically could be 20% or higher. Researchers must compute and report the class-specific FDR for variant peptides, beyond the global FDR. This requires that the search output be partitioned into variant and reference peptide classes before FDR calculation.

Site-Level Quality Control

Beyond peptide-level FDR, proteogenomic analysis requires site-level quality control. A peptide containing a variant amino acid may be correctly identified as a peptide, but the specific site of the variation may be incorrectly localized. This is analogous to the problem of post-translational modification site localization, where the mass shift can be placed at multiple possible sites within the peptide.

One approach to site-level quality control is unrestrictive mass shift relocalization. This method takes the mass shift responsible for the putative SAV and relocates it to alternative positions within the peptide to search for alternative interpretations [<a href="#ref-5">5</a>]. If the mass shift can be explained equally well by a modification at a different site, the confidence in the variant site assignment is reduced. This approach was implemented in the SAVControl tool, which was validated on three colorectal adenocarcinoma cell line datasets with genomic sequencing evidence and tested on a colorectal cancer dataset from The Cancer Genome Atlas [<a href="#ref-5">5</a>]. The results showed that SAVControl effectively removes false detections of SAVs [<a href="#ref-5">5</a>].

For researchers, the practical step is to run site-localization analysis on all putative variant peptides. This analysis should be recorded for each variant site, and variants with ambiguous site localization should be downgraded in confidence or removed from the final list.

The Role of Search Engine Choice

Search engine choice has a measurable effect on false discovery in proteogenomic searches. In the DIA-MS assessment, Fragpipe-MSFragger demonstrated the most conservative and effective performance with the lowest false discovery match ratio, while other pipelines such as DIA-NN and Spectronaut were also evaluated [<a href="#ref-4">4</a>]. The study also found that integrating DIA and data-dependent acquisition (DDA) MS data search outputs enhanced SAAV peptide identification, with a lower FDR observed in DDA searches [<a href="#ref-4">4</a>].

This finding has a practical implication for workflow design. Researchers who rely on a single search engine and a single acquisition mode may miss true variants or report false ones. Running multiple search engines and integrating results across acquisition modes provides a more robust basis for variant identification. The integration step should be designed so that a variant is reported only if it is identified by more than one search engine or in more than one acquisition mode.

Manual Spectrum Inspection as a Quality Control Step

Automated FDR estimation and site-localization are necessary but not sufficient for confident variant identification. Manual spectrum inspection remains an essential quality control step, particularly for variant peptides that will be reported as novel findings or used for downstream validation.

What to Inspect in a Spectrum

Manual inspection of a peptide-spectrum match requires examining the annotated spectrum for the variant peptide. The key features to check are:

1. Coverage of the fragment ion series. A confident spectrum should have a series of b-ions and y-ions that cover most of the peptide sequence, including the region around the variant site.

1. The presence of fragment ions that specifically support the variant amino acid. If the variant is a substitution, there should be fragment ions that differ in mass from the wild-type peptide by the mass difference of the substitution. The absence of such ions is a warning sign.

1. The absence of unexplained dominant peaks. If the spectrum contains intense peaks that are not assigned to the peptide sequence, this may indicate that the spectrum is a mixture of two co-eluting peptides or that the assignment is incorrect.

1. The mass error of the assigned fragment ions. Fragment ions should match the theoretical masses within the mass accuracy of the instrument. Systematic mass errors across the spectrum suggest a mis-assignment.

The Limitations of Manual Inspection

Manual inspection is time-consuming and not uniform across laboratories [<a href="#ref-2">2</a>]. Different researchers may apply different criteria for what constitutes a confident spectrum, leading to inconsistent results. This lack of uniformity is a recognized problem in the field [<a href="#ref-2">2</a>].

To address this, researchers should establish a written standard operating procedure for manual spectrum inspection before beginning the review process. The procedure should specify the minimum number of fragment ions required, the required coverage around the variant site, and the acceptable mass error. All manual inspection decisions should be recorded, including the rationale for accepting or rejecting each spectrum.

Assisted Manual Validation

Software tools can support manual validation by providing annotations and prioritization. The PgxSAVy tool, for example, provides a framework for quality control and annotation of variant peptides based on variant quality, isobaric masses, and disease annotation [<a href="#ref-2">2</a>]. It also annotates variants based on pathogenicity and provides support for assisted manual validation [<a href="#ref-2">2</a>].

The practical benefit of such tools is that they prioritize the most promising variant peptides for manual inspection, reducing the burden of reviewing every spectrum. Researchers should use these tools to triage their variant lists, focusing manual inspection effort on variants that pass automated quality filters and have biological or clinical relevance.

Orthogonal Validation Using Synthetic Peptides and Targeted Mass Spectrometry

Manual spectrum inspection and FDR control can reduce false positives, but they cannot eliminate them. The gold standard for confirming a variant peptide identification is orthogonal validation using synthetic peptides and targeted mass spectrometry.

Stable Isotope Dilution and Parallel Reaction Monitoring

Stable isotope dilution with parallel reaction monitoring (SID-PRM) is a targeted mass spectrometry approach that provides high-confidence confirmation of peptide identifications. In this approach, a synthetic version of the putative variant peptide is synthesized with stable isotope-labeled amino acids. The labeled synthetic peptide has a known mass shift relative to the endogenous peptide. The sample is then analyzed by PRM, which targets the specific mass-to-charge ratios of both the endogenous and labeled peptides.

If the endogenous peptide is truly present in the sample, it will elute at the same retention time as the labeled synthetic peptide and produce fragment ions with the expected mass differences. The co-elution of the endogenous and labeled peptides, along with matching fragment ion patterns, provides strong evidence that the variant peptide is genuinely present.

In the DIA-MS assessment study, SID-PRM validation confirmed the SAAV peptides identified by DIA-MS and DDA-MS searches, highlighting the reliability of the approach [<a href="#ref-4">4</a>]. This validation step is particularly important for variants that will be reported as novel findings or used for clinical or biological interpretation.

When to Use Orthogonal Validation

Orthogonal validation is resource-intensive and should be prioritized for the most important findings. The following criteria should be used to select variants for validation:

1. Variants with biological or clinical significance, such as known disease-associated mutations or predicted neoantigens.

1. Variants that pass all automated quality filters but have not been previously reported.

1. Variants that will be used for quantitative comparisons between conditions.

1. Variants identified in only one search engine or one acquisition mode.

The number of variants selected for validation should be recorded, along with the validation success rate. A low validation success rate indicates that the upstream filtering is too permissive and should be made more stringent.

The Cost of Validation

Synthetic peptide synthesis and targeted mass spectrometry analysis have associated costs in time and money. The decision to validate must balance the cost against the importance of the finding. For large-scale discovery studies, it may be practical to validate only a subset of variants. For focused studies of specific variants, validation of all reported variants may be appropriate.

Researchers should document the validation strategy in the study methods, including the number of variants selected, the selection criteria, and the validation results. This documentation allows readers to assess the confidence of the reported findings.

Data-Independent Acquisition for Proteogenomic Searches

Data-independent acquisition (DIA) mass spectrometry has become an important tool for proteogenomic analysis. DIA-MS provides reproducible and comprehensive coverage of the proteome, and it can identify a substantial proportion of SAAV peptides [<a href="#ref-4">4</a>]. The assessment of DIA-MS for SAAV identification in HeLa cells demonstrated that DIA-MS searches maintained consistent identification of SAAV peptides despite varying sizes of the entrapment database [<a href="#ref-4">4</a>].

Advantages of DIA for Variant Identification

DIA-MS acquires fragment ion spectra for all precursor ions within a defined mass range, instead of selecting individual precursors for fragmentation as in DDA. This comprehensive acquisition provides several advantages for proteogenomics:

1. Reproducibility. DIA-MS is highly reproducible across runs, which is important for comparing variant identifications across samples.

1. Completeness. DIA-MS provides broader coverage of the proteome, increasing the chance of detecting low-abundance variant peptides.

1. Consistency. The consistent identification of SAAV peptides despite varying entrapment database sizes suggests that DIA-MS results are robust to database composition [<a href="#ref-4">4</a>].

Combining DIA and DDA

The integration of DIA and DDA data search outputs can enhance SAAV peptide identification, with a lower FDR observed in DDA searches [<a href="#ref-4">4</a>]. This finding suggests that the two acquisition modes provide complementary information. DDA searches may provide more confident identifications for individual peptides, while DIA provides broader coverage.

For researchers with access to both acquisition modes, running both and integrating the results provides a more complete and confident variant list. The integration should be performed at the peptide level, with a variant reported only if it is identified in at least one mode with high confidence and confirmed in the other mode or by manual inspection.

Search Engine Pipelines for DIA Data

The choice of search engine pipeline for DIA data affects the false discovery rate. In the HeLa cell assessment, Fragpipe-MSFragger demonstrated the most conservative and effective performance with the lowest false discovery match ratio [<a href="#ref-4">4</a>]. Other pipelines, including DIA-NN and Spectronaut, were also evaluated [<a href="#ref-4">4</a>].

Researchers should test multiple search engine pipelines on their DIA data and compare the results. The pipeline that produces the lowest false discovery match ratio while maintaining acceptable sensitivity should be selected for routine use. The comparison should be repeated when new versions of the search engines are released.

At a Glance: Confidence Assessment Framework

The following table summarizes the key decisions and actions for assessing confidence in proteogenomic search results.

Practical Workflow for Proteogenomic Confidence Assessment

The following workflow provides a step-by-step procedure for evaluating proteogenomic search results. This workflow is designed to be implemented in any laboratory performing proteogenomic analysis.

Step 1: Document the Database Construction

Before any search is run, the database construction must be fully documented. Record the reference proteome version, the source of variant sequences (WES data, public mutation database, or customized cell line database), the filtering criteria applied to variants, and the final database size. Include the entrapment sequences in the database and record their identifiers.

The database construction decisions should be made with the understanding that database complexity directly affects false discovery. A database that includes every possible variant from a public database without filtering will produce more false positives than a filtered database. The filtering criteria should be based on variant quality scores, predicted protein-coding effect, and allele frequency.

Step 2: Run Searches with Multiple Engines

Run the proteogenomic search with at least two search engines. For DIA data, consider DIA-NN, Spectronaut, and Fragpipe-MSFragger [<a href="#ref-4">4</a>]. For DDA data, use the search engines appropriate for the data format. Record the version numbers of all search engines and the search parameters used.

The search parameters should be consistent across engines where possible, including precursor mass tolerance, fragment mass tolerance, and enzyme specificity. Differences in search parameters between engines can confound the comparison of results.

Step 3: Compute Class-Specific FDR

After the searches are complete, compute the FDR separately for variant peptides and reference peptides. The class-specific FDR for variant peptides is the critical metric. If the variant class-specific FDR exceeds the acceptable threshold (typically 1% or 5%, depending on the study), the search parameters must be made more stringent.

The FDR calculation should be based on the target-decoy approach, with decoy sequences generated by reversing or shuffling the target sequences. The decoy database should include decoys for both reference and variant sequences.

Step 4: Run Site Localization Analysis

For all variant PSMs that pass the FDR threshold, run site localization analysis using unrestrictive mass shift relocalization [<a href="#ref-5">5</a>]. This analysis determines whether the mass shift responsible for the variant can be explained by an alternative site within the peptide.

Record the site localization score for each variant. Variants with ambiguous site localization should be flagged for manual inspection or removed from the final list.

Step 5: Apply Automated Quality Filters

Apply additional automated filters to the variant list. These filters may include:

1. Minimum peptide length and maximum missed cleavage count.

1. Minimum number of fragment ions matched.

1. Maximum mass error for fragment ions.

1. Removal of variants that are isobaric with common post-translational modifications.

The PgxSAVy tool provides a framework for this filtering, integrating intensity, variant event information, and search result features to discriminate correct variant peptides [<a href="#ref-2">2</a>]. The tool classifies variants as confident, semi-confident, or doubtful [<a href="#ref-2">2</a>].

Step 6: Perform Manual Spectrum Inspection

Prioritize the variant PSMs for manual inspection. The prioritization should be based on the automated quality scores and the biological or clinical relevance of the variant. Apply the written standard operating procedure for spectrum inspection to each prioritized spectrum.

Record the inspection decision for each spectrum, including the reason for acceptance or rejection. The manual inspection results should be reported in the study methods.

Step 7: Select Variants for Orthogonal Validation

Select a subset of variants for orthogonal validation using SID-PRM. The selection criteria should be documented. Synthesize stable isotope-labeled peptides for the selected variants and run PRM analysis.

Record the validation results, including the number of variants confirmed and the number that could not be confirmed. A low confirmation rate indicates that the upstream filtering is too permissive.

Step 8: Report the Results with Confidence Levels

Report the final variant list with confidence levels for each variant. The confidence levels should reflect the evidence supporting each identification, including the FDR, site localization score, manual inspection result, and orthogonal validation status.

The reporting should include the total number of variant PSMs identified, the number passing each filtering stage, and the final number of confident variants. This information allows readers to assess the overall false discovery risk in the study.

Records and Measurements for Proteogenomic Quality Control

Maintaining detailed records is essential for reproducible and trustworthy proteogenomic analysis. The following records should be maintained for every proteogenomic study.

Database Records

The database records should include the reference proteome version and download date, the variant source and version, the filtering criteria applied, the number of variant sequences included, the number of entrapment sequences included, and the final database size. Any changes to the database during the analysis should be recorded with dates and reasons.

Search Records

The search records should include the search engine names and versions, the search parameters used, the date and time of each search, and the computational resources used. The raw search output files should be archived.

FDR Records

The FDR records should include the global FDR, the class-specific FDR for variant and reference peptides, the number of target and decoy identifications at each threshold, and the method used for decoy generation.

Validation Records

The validation records should include the list of variants selected for validation, the selection criteria, the synthetic peptide sequences and labels, the PRM acquisition parameters, and the validation results for each variant.

Sample Records

The sample records should include the sample source, the sample preparation method, the mass spectrometry instrument and acquisition parameters, and the raw data file names. These records are necessary for reproducing the analysis and for assessing the comparability of results across studies.

Common Failure Patterns in Proteogenomic Analysis

Recognizing common failure patterns can help researchers identify problems in their analysis before they lead to false reports.

Pattern 1: Overly Permissive FDR Thresholds

A common failure is the use of a global FDR threshold without computing the class-specific FDR for variant peptides. This can result in a variant list with a much higher false discovery rate than reported. The fix is to compute and report class-specific FDR.

Pattern 2: Ignoring Site Localization

Variant peptides may be correctly identified as peptides, but the variant site may be incorrectly localized. Reporting a variant at the wrong site is a false positive even if the peptide sequence is correct. The fix is to run site localization analysis and flag ambiguous sites.

Pattern 3: Single Search Engine Reliance

Relying on a single search engine can produce results that are specific to that engine's scoring algorithm. Variants identified by only one engine are less reliable than those identified by multiple engines. The fix is to run multiple engines and integrate results.

Pattern 4: Skipping Manual Inspection

Automated filters cannot catch all false positives. Skipping manual inspection of prioritized spectra can result in the reporting of spectra that do not support the variant assignment. The fix is to implement a written manual inspection procedure.

Pattern 5: No Orthogonal Validation

Reporting novel variants without orthogonal validation leaves the identifications unconfirmed. The fix is to validate the most important variants using SID-PRM or another targeted approach.

Pattern 6: Uncontrolled Database Complexity

Using an unfiltered database with all possible variants from a public database increases the search space and can increase false discoveries. The fix is to filter variants by quality and predicted effect and to include entrapment sequences as a control.

Pattern 7: Inconsistent Manual Review

Manual inspection that is not guided by a written standard operating procedure produces inconsistent results across laboratories and across reviewers within a laboratory [<a href="#ref-2">2</a>]. The fix is to establish and follow a written procedure.

Limitations of Current Approaches

The approaches described in this article have limitations that researchers should understand.

FDR Estimation Limitations

Class-specific FDR estimation requires that the variant peptide class contains enough identifications for reliable estimation. In studies with few variant identifications, the FDR estimate may be unstable. The target-decoy approach assumes that decoy sequences accurately model false target identifications, but this assumption may not hold for variant peptides, which are structurally similar to reference peptides.

Manual Inspection Limitations

Manual spectrum inspection is subjective and time-consuming. Different reviewers may reach different conclusions about the same spectrum. The lack of uniform criteria across laboratories is a recognized problem [<a href="#ref-2">2</a>]. While software tools can support manual inspection, they cannot replace the expert judgment required for difficult cases.

Orthogonal Validation Limitations

SID-PRM validation requires the synthesis of labeled peptides, which is expensive and time-consuming. The validation can only confirm the presence of the peptide in the sample, it cannot confirm that the peptide is translated from the predicted genomic variant. The validation also requires that the peptide is detectable by PRM, which may not be the case for low-abundance peptides.

Database Limitations

The completeness of the variant database limits the sensitivity of the search. Variants that are not in the database cannot be identified. WES-based databases may miss variants due to sequencing depth limitations and sample heterogeneity [<a href="#ref-3">3</a>]. Public mutation databases may not contain variants that are specific to the sample being studied.

Search Engine Limitations

Different search engines have different strengths and weaknesses. The performance of a search engine for one type of data or one type of variant may not generalize to other contexts. The search engine landscape changes rapidly, and new versions may perform differently than older versions.

Safety and Reproducibility Context

Proteogenomic analysis is a computational discipline, and the primary risks are to the validity of scientific conclusions instead of to physical safety. However, there are important considerations for the responsible conduct of research.

Reproducibility Standards

The Galaxy Training Network (GTN) hosts peer-reviewed, browser-accessible tutorials that walk researchers through reproducible analysis pipelines.

Researchers should follow these standards and document their analysis in sufficient detail for reproduction. The raw mass spectrometry data should be deposited in public repositories, and the analysis code and parameters should be made available.

Training and Skill Development

Proteogenomic analysis requires skills in mass spectrometry, bioinformatics, and statistics. The EMBL-EBI Training program provides learning pathways for bioinformatics and data-resource training [<a href="#ref-6">6</a>]. The Carpentries lessons provide foundational computing, data, shell, Git, and programming training [<a href="#ref-7">7</a>]. The NCBI provides official descriptions of databases, search systems, sequence resources, and analysis services [<a href="#ref-8">8</a>].

Researchers who are new to proteogenomics should complete training in these areas before attempting complex analyses. The training should include hands-on practice with the specific tools and databases used in the analysis.

Data Management

The data generated by proteogenomic analysis includes raw mass spectrometry files, search output files, and analysis results. These data should be managed according to the principles of good data management, including regular backups, version control, and documentation. The nf-core documentation provides guidance on reproducible workflow configuration that includes data management practices [<a href="#ref-9">9</a>].

Escalation Criteria

Researchers should escalate to a supervisor, collaborator, or expert when they encounter situations that exceed their expertise. The following situations warrant escalation:

1. The class-specific FDR for variant peptides cannot be reduced to an acceptable level with any reasonable search parameter adjustment.

1. Manual spectrum inspection reveals systematic problems with the search results that cannot be explained by known limitations of the search engine.

1. Orthogonal validation fails to confirm a substantial proportion of the variants that passed all automated filters.

1. The results of different search engines are highly discordant, and the source of the discordance cannot be identified.

1. A variant with potential clinical significance is identified, and the researcher is uncertain about the appropriate validation strategy.

1. The analysis requires computational resources that exceed the capacity of the available infrastructure.

1. The researcher is uncertain about the appropriate interpretation of the results in the context of the biological question being addressed.

Frequently Asked Questions

What is the difference between global FDR and class-specific FDR in proteogenomics?

Global FDR is calculated across all peptide identifications in a search, treating reference peptides and variant peptides as a single class. Class-specific FDR is calculated separately for variant peptides and reference peptides. In proteogenomic searches, the class-specific FDR for variant peptides is typically much higher than the global FDR because variant peptides are rare and more difficult to identify correctly. Reporting only the global FDR can mislead readers into thinking that the variant identifications are more reliable than they actually are. Researchers should compute and report both values.

How do entrapment databases help control false positives?

Entrapment databases contain sequences that are known to be absent from the sample, such as proteins from a distantly related organism. When a search returns identifications from the entrapment sequences, this indicates that the search is producing false positives. The number of entrapment identifications provides a direct measure of the false discovery rate. In the assessment of DIA-MS for SAAV identification, entrapment databases were used to evaluate the false discovery match ratio across different search engine pipelines [<a href="#ref-4">4</a>]. Researchers should include entrapment sequences in every proteogenomic search and record the entrapment identification rate.

Why is site localization important for variant peptide identification?

A variant peptide may be correctly identified as a peptide, but the specific site of the amino acid variation may be incorrectly assigned. This is analogous to the problem of post-translational modification site localization. If the mass shift responsible for the variant can be explained equally well by a modification at a different site, the confidence in the variant site assignment is reduced. Unrestrictive mass shift relocalization is a method for testing alternative site assignments [<a href="#ref-5">5</a>]. Variants with ambiguous site localization should be flagged or removed from the final list.

When should I use synthetic peptides for validation?

Synthetic peptide validation should be used for the most important findings in a proteogenomic study. This includes variants with biological or clinical significance, variants that pass all automated filters but have not been previously reported, variants that will be used for quantitative comparisons, and variants identified by only one search engine or acquisition mode. Stable isotope dilution with parallel reaction monitoring (SID-PRM) provides high-confidence confirmation by comparing the retention time and fragment ion pattern of the endogenous peptide with a labeled synthetic standard [<a href="#ref-4">4</a>]. The validation strategy should be documented in the study methods.

How do I choose between a WES-based database and a public mutation database?

The choice depends on the research question and the sample being studied. A WES-based database is specific to the sample but may miss variants due to sequencing depth limitations and sample heterogeneity [<a href="#ref-3">3</a>]. A public mutation database such as COSMIC provides broader coverage but includes variants that may not be present in the sample. In a study of HLA-I immunopeptidome identification, the COSMIC-based database identified more mutant peptides than the WES-based database [<a href="#ref-3">3</a>]. Researchers should consider the completeness of their WES data and the relevance of public database variants to their sample when making this decision.

What is the role of manual spectrum inspection in proteogenomic analysis?

Manual spectrum inspection is a quality control step that cannot be replaced by automated filters. A trained researcher examines the annotated spectrum for each variant peptide to verify that the fragment ion series supports the assignment, that the variant site is supported by specific fragment ions, and that there are no unexplained dominant peaks. Manual inspection is time-consuming and not uniform across laboratories [<a href="#ref-2">2</a>]. Researchers should establish a written standard operating procedure for manual inspection and record all inspection decisions.

How do DIA and DDA compare for variant peptide identification?

DIA-MS provides reproducible and comprehensive coverage of the proteome and can identify a substantial proportion of SAAV peptides [<a href="#ref-4">4</a>]. DDA searches tend to have a lower FDR for individual peptide identifications [<a href="#ref-4">4</a>]. Integrating DIA and DDA data search outputs can enhance SAAV peptide identification [<a href="#ref-4">4</a>]. Researchers with access to both acquisition modes should consider running both and integrating the results.

What should I do if my variant validation rate is low?

A low validation rate indicates that the upstream filtering is too permissive. The first step is to review the class-specific FDR for variant peptides and make the FDR threshold more stringent. The second step is to review the site localization results and remove variants with ambiguous site assignments. The third step is to review the manual inspection results and ensure that the inspection criteria are being applied consistently. If the validation rate remains low after these steps, the search parameters and database construction should be reviewed.

Related Bioinformatics Guides

• Spatial Transcriptomics Study Design: Key Considerations for Robust Results
• Object Detection in Medical Images: Algorithms, Datasets, and Validation Strategies
• RNA-Seq vs qPCR: Validation and Comparison
• How to Interpret Gene Set Enrichment Analysis Results
• 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.