# From Raw Files to Peptide Identifications: A Step-by-Step Guide to Mass Spectrometry Data Preprocessing in Proteomics

Mass spectrometry based proteomics produces raw instrument files that require multiple computational preprocessing steps before peptide identifications can be trusted. This article walks through the complete pipeline from spectral acquisition through peak picking, centroiding, mzML conversion, quality control, and database searching, with attention to the decisions that determine whether downstream protein inferences are biologically meaningful. The intended reader is a graduate student, postdoctoral researcher, or laboratory professional who has generated or received raw mass spectrometry files and needs a practical path to peptide identifications without getting lost in the fragmented software landscape.

## Scope of the Preprocessing Pipeline

The computational workflow described here covers the transformation of raw mass spectrometry output into a list of confidently identified peptide sequences. This pipeline sits between two boundaries. On the upstream side, sample preparation, chromatography, and instrument acquisition parameters determine what the raw files contain. On the downstream side, protein inference, quantification, and statistical analysis convert peptide identifications into biological conclusions. Preprocessing occupies the middle ground and includes spectral quality assessment, peak detection, charge state assignment, mass calibration, data format conversion, and database searching with false discovery rate control.

Researchers new to the field often assume that peptide identification begins with database searching. In practice, the quality of identifications depends heavily on what happens before the search engine sees the data. Poorly calibrated spectra, incorrectly centroided peaks, or files converted with incompatible settings will produce misleading results even when the search algorithm is well chosen. The preprocessing decisions described here are therefore analytical steps that determine the reliability of every subsequent conclusion.

The workflow applies to data dependent acquisition (DDA), which remains the most common mode for discovery proteomics. Data independent acquisition (DIA) uses a different preprocessing logic and is mentioned where relevant, but the step by step path here focuses on DDA because it is the entry point for most new practitioners. The tools referenced throughout are open source or freely available, and the principles transfer across vendor platforms.

## Raw Data Formats and the Need for Standardization

Every mass spectrometer vendor stores raw data in a proprietary format. Thermo instruments produce .raw files, Bruker instruments produce .d folders, and SCIEX instruments produce .wiff files. These formats encode the measured spectra, instrument settings, acquisition timing, and vendor specific metadata. The proprietary nature of these formats creates a practical problem for the proteomics community because downstream tools must be able to read the data regardless of which instrument produced it.

The solution adopted by the field is the mzML format, an open standard developed by the Proteomics Standards Initiative. Converting vendor raw files to mzML creates a vendor neutral representation of the spectra that any analysis tool can read. This conversion is the first major preprocessing step and should be performed with care because the conversion process can introduce artifacts if settings are chosen poorly.

The [EMBL-EBI training portal](https://www.ebi.ac.uk/training) provides structured learning pathways for bioinformatics data resources and practical analysis education, including material on mass spectrometry data standards and file formats. New practitioners should consult these materials to understand the relationship between vendor formats and open standards before beginning their first conversion.

Conversion tools include msConvert from the ProteoWizard toolkit, which is the most widely used option. The tool reads vendor files and writes mzML output with configurable options for peak picking, centroiding, and data compression. The choice between preserving profile data or converting to centroided data during conversion is a critical decision that affects all downstream analysis.

## Peak Picking and Centroiding

Raw mass spectrometry data are acquired in profile mode, meaning each peak is represented by many data points across its mass to charge range. Profile data preserve the full shape of the peak, including the rise, apex, and fall. This representation is information rich but computationally heavy and poorly suited to database searching, which expects a discrete list of peak positions and intensities.

Centroiding, also called peak picking, reduces each profile peak to a single data point representing the monoisotopic mass and intensity. The process involves detecting the peak, determining its apex, calculating the area or height, and assigning a centroid position. The quality of this step directly affects mass accuracy in the downstream search because a poorly centered peak will produce a measured mass that deviates from the true value.

The decision of when to centroid matters. Some pipelines centroid during conversion with msConvert, while others preserve profile data and centroid later with specialized tools. The choice depends on the downstream search engine and the preference of the analysis pipeline. What matters is consistency. Mixing centroided and profile spectra in a single search produces unreliable results because search engines interpret the two data types differently.

The [Bioconductor project](https://bioconductor.org/) hosts multiple R packages for mass spectrometry data processing, including packages that perform peak picking and centroiding on mzML files. These packages provide programmatic control over the peak detection parameters, which is valuable for researchers who need to optimize the process for specific instrument types or sample complexities.

## Mass Calibration and Its Role in Identification Confidence

Mass accuracy is a primary determinant of peptide identification confidence. Modern instruments achieve parts per million mass accuracy, and search engines use this accuracy to restrict the number of candidate peptides considered for each spectrum. A systematic mass offset across the m/z range will cause true matches to be missed and can inflate false identifications if the search tolerance is widened to compensate.

Calibration errors in tandem mass spectra are common and often small, yet these small errors can degrade interpretation and annotation quality. A tool called SpectiCal addresses this problem by reading mzML files from data dependent acquisition experiments, computing m/z calibrations for each file prior to identification analysis, and using known low mass ions as calibration references. The tool produces information about frequently observed peaks and their explanations, and it can generate new calibrated mzML files using the computed calibration coefficients. Testing on public data sets produced a table of commonly observed low mass ions and their identifications, with information about the number of runs in which each peak appears and the percentage of spectra containing it. SpectiCal can also examine MS runs for artifacts that might hinder downstream analysis, making it useful both for calibration and for quality assessment before database searching.

The [SpectiCal publication](https://pubmed.ncbi.nlm.nih.gov/38538550) describes the tool and its validation on public data sets. Researchers who observe systematic mass errors in their searches should consider calibration as a preprocessing step instead of simply widening the search tolerance. Widening tolerance increases the number of candidate peptides and therefore increases the false discovery rate, which is a poor trade for correcting a correctable systematic error.

## Quality Control Before Database Searching

Quality control should occur at multiple points in the preprocessing pipeline, including after identifications are complete. Early quality control catches instrument problems, sample preparation failures, and acquisition errors before they waste hours of database searching time.

At the raw file level, quality indicators include total ion current, number of MS2 spectra acquired, precursor intensity distribution, and chromatographic peak shape. A run with very few MS2 spectra may indicate that the instrument failed to trigger fragmentation, which could reflect a problem with precursor selection settings or with sample concentration. A run with unusual total ion current patterns may indicate a chromatography problem or an air leak in the instrument.

The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow training and analysis tutorials that include quality control steps for mass spectrometry data. These tutorials are useful for researchers who want to see complete workflows with quality assessment built in, instead of assembling individual tools without guidance.

After conversion to mzML, additional quality checks become possible. The number of peaks per spectrum, the distribution of precursor charge states, and the mass error distribution from an initial search all provide diagnostic information. A preliminary search against a small database can reveal whether the data are interpretable before committing to a full search.

## Database Searching Fundamentals

Database searching compares each tandem mass spectrum against theoretical spectra generated from a protein sequence database. The search engine fragments each candidate peptide in silico, predicts the resulting fragment ions, and scores the match between predicted and observed spectra. The output is a list of peptide spectrum matches (PSMs) with associated scores.

The choice of protein database is a preprocessing decision that has major consequences. A database that is too small may miss true identifications. A database that is too large increases the search space and can reduce statistical power. For well characterized organisms, the reference proteome from a major database is usually appropriate. For less characterized organisms or for samples that may contain contaminants, the database choice requires more thought.

The [NCBI data resources](https://www.ncbi.nlm.nih.gov/) provide official descriptions of sequence databases, search systems, and analysis services that are relevant to database selection. Researchers should understand the difference between a reference proteome, a complete proteome, and a database that includes isoforms, because this choice affects both sensitivity and specificity.

Common search engines include MaxQuant with its built in Andromeda search, MSFragger, and the OpenMS suite. Each has strengths and weaknesses, and the choice often depends on the specific application. MaxQuant is popular for label free quantification because it integrates identification and quantification in a single pipeline. MSFragger is known for its speed and for its ability to search open mass tolerances, which is useful for finding unexpected modifications.

## False Discovery Rate Control

Peptide identifications require statistical validation because search scores alone do not distinguish true from false matches. The standard approach is the target decoy strategy, in which the search is performed against a database containing both real protein sequences and reversed or shuffled decoy sequences. Matches to decoy sequences provide an estimate of the false discovery rate.

The false discovery rate (FDR) is typically controlled at 1 percent for peptide identifications and 1 percent for protein identifications, though the specific thresholds depend on the study requirements. The target decoy approach assumes that false matches are equally likely to hit target or decoy sequences, which allows the number of decoy hits to estimate the number of false target hits.

The [nf-core documentation](https://nf-co.re/docs) describes community pipeline standards for reproducible workflows, including pipelines that implement target decoy searching and FDR control. These pipelines provide a template for researchers who want to ensure their analysis follows community accepted practices.

FDR control is not a single step but a series of decisions. The search engine assigns scores, the FDR is calculated at the PSM level, then at the peptide level, then at the protein level. Each level has its own statistical considerations. The researcher must decide whether to filter at the PSM level before protein inference or to carry all PSMs through and filter at the protein level. These decisions affect the final list of identifications.

## Protein Inference and the Parsimony Principle

Peptide identifications must be assembled into protein identifications, and this step introduces ambiguity. A peptide sequence may match multiple proteins, either because the proteins are isoforms, share domains, or are homologous. The parsimony principle states that the minimal set of proteins that explains all observed peptides should be reported.

Protein inference is a source of both confusion and error in proteomics. Researchers who are new to the field often expect a one to one mapping between peptides and proteins, but the reality is more complex. A single peptide can match many proteins, and a protein may be identified by a single peptide or by many peptides. The distinction between protein groups, which are sets of proteins that cannot be distinguished by the observed peptides, is essential for honest reporting.

The [Cross-ID publication](https://pubmed.ncbi.nlm.nih.gov/30575379) illustrates the complexity of protein level analysis in the context of cross linking mass spectrometry, where protein interaction networks are the analytical target. The tool described there supports grouping, spectral viewing, gene ontology enrichment, and post-translational modification visualization, demonstrating that protein level interpretation requires dedicated software beyond the basic search output.

For standard proteomics, the search engine output typically includes protein groups with associated peptides and scores. The researcher must decide how to handle proteins identified by a single peptide, which are less confident than proteins identified by multiple peptides. Some studies require a minimum of two peptides per protein, while others accept single peptide identifications with additional evidence.

## Quantification Considerations in the Preprocessing Pipeline

Although the focus of this article is peptide identification, quantification considerations affect preprocessing decisions. Label free quantification requires consistent peak detection and alignment across runs. Isobaric labeling approaches such as TMT or iTRAQ require reporter ion extraction from the MS2 or MS3 spectra. Each quantification strategy imposes its own preprocessing requirements.

For label free quantification, the preprocessing pipeline must preserve intensity information accurately. Centroiding parameters affect peak area calculations, and the choice of peak picking algorithm influences the reproducibility of quantification across runs. The [MS-DAP publication](https://pubmed.ncbi.nlm.nih.gov/36541440) describes a downstream analysis pipeline that integrates normalization and statistical analysis algorithms and emphasizes the importance of transparent and reproducible proteome science with standardized quality reporting. The systematic evaluation in that study revealed key differences among normalization and statistical analysis methods, with commonly used approaches based on moderated t statistics outperformed by more recent statistical models.

For isobaric labeling, the preprocessing pipeline must extract reporter ion intensities from the spectra. This extraction requires accurate peak picking in the low mass region of the MS2 or MS3 spectra, where the reporter ions appear. The [SpectiCal tool](https://pubmed.ncbi.nlm.nih.gov/38538550) is relevant here because it focuses on low mass ions and can identify artifacts in this region that might interfere with reporter ion quantification.

Researchers should decide on their quantification strategy before finalizing the preprocessing pipeline, because the choices are interconnected. A pipeline optimized for identification only may not preserve the information needed for quantification, and redoing the preprocessing after the fact is costly.

## At a Glance: Preprocessing Decision Points

The table below summarizes the major preprocessing stages, the key decisions at each stage, and the records that should be maintained for reproducibility.

| Pipeline Stage | Primary Decision | Records to Maintain |
| --- | --- | --- |
| Format conversion | Centroid during conversion or preserve profile data | Conversion tool version, input and output file names, peak picking settings |
| Mass calibration | Apply per file calibration or rely on instrument calibration | Calibration coefficients, number of low mass ions used, calibrated file names |
| Database selection | Reference proteome, complete proteome, or custom database with contaminants | Database name, version, download date, number of sequences |
| Search configuration | Precursor and fragment mass tolerances, enzyme specificity, variable modifications | Search engine version, parameter file, FDR threshold |
| FDR filtering | PSM level, peptide level, or protein level threshold | Score distributions, decoy hit counts, final identification counts |

## Common Failure Patterns in Preprocessing

Several failure patterns recur in proteomics preprocessing, and recognizing them early saves substantial time and effort.

The first pattern is inconsistent centroiding. When some spectra in a file are centroided and others are not, search engines produce erratic results. This situation arises when conversion settings change mid project or when files from different instruments are combined without checking the data format. The solution is to verify the data format of every file before searching and to standardize the conversion settings across the entire project.

The second pattern is incorrect charge state assignment. Precursor charge state affects the calculation of precursor mass and the interpretation of fragment ions. Modern search engines can determine charge state from the data, but errors occur when the precursor isolation window includes multiple species or when the charge state is ambiguous. Checking the charge state distribution across the dataset can reveal systematic problems.

The third pattern is database mismatch. Searching a sample from one organism against the wrong database produces few identifications and a high proportion of decoy hits. This problem is common when samples contain multiple organisms, such as host and pathogen, or when the sample is contaminated with common laboratory contaminants. Including a contaminant database and checking the species distribution of identifications can diagnose this problem.

The fourth pattern is mass calibration drift. When the instrument calibration drifts during a long acquisition, the mass errors increase over the course of the run. This pattern appears as a systematic increase in mass error with retention time. The [SpectiCal approach](https://pubmed.ncbi.nlm.nih.gov/38538550) of computing per file calibrations based on known low mass ions addresses this problem directly.

The fifth pattern is overfiltering or underfiltering at the FDR step. Overfiltering removes true identifications to achieve an artificially low FDR, while underfiltering reports false identifications as real. Both problems stem from misunderstanding the FDR calculation or from applying thresholds without examining the score distributions.

## Reproducibility and Workflow Management

Reproducibility in proteomics preprocessing requires more than documenting the tools used. The exact parameters, database versions, and software versions must be recorded, and the workflow must be executable by another researcher. Containerization and workflow management systems address this need.

The [nf-core documentation](https://nf-co.re/docs) describes community standards for reproducible workflows, including pipeline structure, configuration, and usage. These standards ensure that a pipeline can be run by different researchers on different systems with consistent results. The documentation covers best practices for workflow design that apply to proteomics preprocessing.

The [Galaxy Training Network](https://training.galaxyproject.org/) provides another route to reproducibility through its web based platform, where workflows can be shared and rerun with recorded parameters. This approach is accessible to researchers who do not have command line experience.

The [Carpentries lessons](https://carpentries.org/lessons) cover foundational computing skills including shell, Git, and programming, which are prerequisites for reproducible workflow management. Researchers who invest in these skills are better equipped to manage the complexity of proteomics preprocessing.

Version control is essential for reproducibility. The database version, search engine version, and parameter files must be tracked. A common failure is the inability to reproduce results because the database was updated between searches or because the parameter file was overwritten. Recording the exact database download date and the parameter file contents prevents this problem.

## Practical Implementation Steps

The following steps provide a practical path from raw files to peptide identifications. Each step includes the key decisions and the records that should be kept.

Step one is inventory and inspection. List all raw files in the project and record their acquisition dates, instrument settings, and sample annotations. Inspect the total ion chromatograms and base peak chromatograms for each file to identify runs with obvious problems such as no peaks, unusual baselines, or chromatography failures. Files with severe problems should be flagged for potential exclusion before further processing.

Step two is format conversion. Convert vendor raw files to mzML using msConvert or an equivalent tool. Decide whether to centroid during conversion or preserve profile data. Record the conversion settings, including the peak picking algorithm and the m/z tolerance. Verify that the output files are readable and that the number of spectra in each mzML file matches the expected number from the raw file.

Step three is quality assessment of the converted files. Examine the number of MS2 spectra per file, the precursor charge state distribution, and the precursor intensity distribution. Files with unusually low MS2 counts or unusual charge state distributions should be investigated before proceeding. The [Galaxy Training Network](https://training.galaxyproject.org/) tutorials provide examples of quality assessment workflows.

Step four is mass calibration if needed. If previous searches have shown systematic mass errors, apply a calibration tool such as SpectiCal to compute per file calibrations and generate calibrated mzML files. Record the calibration coefficients and the number of low mass ions used for calibration.

Step five is database selection. Choose the protein database appropriate for the sample organism and the study question. Record the database name, version, download date, and the number of protein sequences. Add common contaminants to the database if the search engine does not include them by default.

Step six is the database search. Configure the search engine with the appropriate precursor mass tolerance, fragment mass tolerance, enzyme specificity, and variable modifications. The tolerance settings should reflect the instrument mass accuracy and the calibration status of the data. Record all search parameters in a parameter file.

Step seven is FDR filtering and quality assessment of identifications. Apply the target decoy FDR threshold appropriate for the study. Examine the number of peptide and protein identifications, the distribution of search scores, and the mass error distribution. Compare the identification numbers across files to identify outlier runs.

Step eight is protein inference and reporting. Assemble peptide identifications into protein groups using the parsimony principle. Record the number of proteins identified, the number of proteins identified by a single peptide, and the protein group sizes. Export the results in a standard format for downstream analysis.

## Records and Measurements

The records kept during preprocessing determine whether the analysis can be audited, reproduced, or extended. The following records should be maintained for every project.

The raw file inventory should include the file name, acquisition date, instrument, acquisition method, sample identifier, and any notes about run quality. This inventory is the foundation of the project record.

The conversion log should record the conversion tool version, the input and output file names, the conversion settings, and any warnings or errors generated during conversion. This log allows the conversion to be reproduced exactly.

The search parameter file should record the database name and version, the search engine version, the precursor and fragment mass tolerances, the enzyme specificity, the variable modifications, and the FDR threshold. This file is the most important record for reproducibility because small parameter changes can produce large differences in results.

The quality control log should record the metrics examined at each quality checkpoint, including the number of MS2 spectra, the charge state distribution, the mass error distribution, and the identification counts. Trends across files can reveal instrument drift or sample degradation that would otherwise go unnoticed.

The [Bioconductor project](https://bioconductor.org/) provides packages for managing and analyzing mass spectrometry data that include functionality for record keeping and quality assessment. These packages can generate standardized reports that document the preprocessing steps and results.

## Limitations of the Preprocessing Pipeline

The preprocessing pipeline described here has limitations that researchers should understand before interpreting results.

The first limitation is that preprocessing cannot compensate for poor sample preparation or poor acquisition. If the sample was degraded, the chromatography was poor, or the instrument was not performing well, no amount of computational processing will produce reliable identifications. The quality control steps in the pipeline can detect these problems, but they cannot fix them.

The second limitation is that database searching only identifies peptides that are present in the database. Novel peptides, unexpected modifications, or peptides from organisms not represented in the database will not be identified. Open search approaches can find unexpected modifications, but they require different preprocessing and search strategies.

The third limitation is that peptide identifications do not directly equal protein identifications. The parsimony principle and protein grouping introduce ambiguity that must be reported honestly. Researchers who ignore this ambiguity and report all proteins matched by any peptide will overstate their results.

The fourth limitation is that the FDR calculation depends on assumptions about the target decoy strategy. If the database is too small, the decoy estimates may be unreliable. If the search space is too large, the FDR may be underestimated. The researcher must understand these assumptions to interpret the FDR correctly.

The fifth limitation is that preprocessing choices affect downstream quantification. A pipeline optimized for identification may not preserve the quantitative information needed for differential expression analysis. The [MS-DAP study](https://pubmed.ncbi.nlm.nih.gov/36541440) demonstrated that the choice of normalization and statistical analysis methods has major effects on the results, and these choices are downstream of the preprocessing decisions described here.

## Specialized Applications and Their Preprocessing Needs

Different proteomics applications impose different preprocessing requirements. Researchers should be aware of these differences before starting a project.

Cross linking mass spectrometry (XL-MS) requires specialized preprocessing because the spectra contain fragments from two linked peptides. The [Cross-ID tool](https://pubmed.ncbi.nlm.nih.gov/30575379) addresses the visualization and analysis of XL-MS data, but the preprocessing must preserve the cross linked peptide information. Standard peak picking and centroiding apply, but the search requires a cross link aware search engine and a database formatted for cross link searching.

Peptidomics, the study of endogenous peptides, requires preprocessing that accounts for the absence of trypsin digestion. The [tsetse fly peptidomics study](https://pubmed.ncbi.nlm.nih.gov/26463237) demonstrates that endogenous peptides can be identified without digestion, but the search parameters must reflect the lack of enzyme specificity. The preprocessing pipeline must also account for the small size of many neuropeptides, which affects the mass range of interest.

Extracellular vesicle proteomics presents challenges related to sample complexity and protein abundance. The [viewpoint article on plasma extracellular vesicles](https://pubmed.ncbi.nlm.nih.gov/40955644) discusses how mass spectrometers can provide reproducible deep coverage of the EV proteome despite sample impurities, and how advanced bioinformatics tools can refine the EV proteome landscape. The preprocessing pipeline for EV samples must handle the complexity of the background proteome and the low abundance of many EV proteins.

Each specialized application requires the researcher to understand how the preprocessing choices affect the specific analytical goal. A pipeline that works well for standard whole cell lysate proteomics may not be appropriate for peptidomics or cross linking studies.

## Professional Escalation Criteria

Researchers should know when to seek help from more experienced colleagues or from core facility staff. The following situations warrant escalation.

If the quality control metrics indicate a systematic instrument problem, such as a consistent mass error drift or a sudden drop in MS2 counts, the instrument should be checked by facility staff before proceeding. Continuing to process data from a malfunctioning instrument wastes time and produces unreliable results.

If the database search produces very few identifications and the mass error distribution is broad, the problem may be in the preprocessing instead of in the sample. An experienced colleague can review the conversion settings and search parameters to identify the issue.

If the FDR filtering produces an unexpected number of identifications, either very high or very low, the search parameters or the database choice should be reviewed. A sudden increase in identifications after a database update may indicate that the new database contains redundant sequences that inflate the identification count.

If the researcher is uncertain about the appropriate database or search parameters for an unfamiliar sample type, consultation with a core facility or an experienced collaborator is appropriate. The [EMBL-EBI training portal](https://www.ebi.ac.uk/training) and the [Galaxy Training Network](https://training.galaxyproject.org/) provide educational materials that can help, but hands on guidance from an experienced practitioner is often necessary.

If the downstream analysis reveals inconsistencies that cannot be traced to the biological question, the preprocessing steps should be audited. The records described above make this audit possible.

## Decision Framework for Selecting Preprocessing Tools and Parameters

New practitioners often struggle with the abundance of available preprocessing tools and the lack of clear guidance on when to use which option. This section provides a practical decision framework based on project characteristics, sample type, and downstream goals. The framework helps researchers make defensible choices without testing every possible combination.

### Project Type and Scale Assessment

The first decision point is project scale and complexity. A small pilot study with fewer than twenty raw files requires a different approach than a clinical cohort study with hundreds of files. For small projects, manual processing with individual tools is feasible and allows close inspection of each step. For large projects, workflow managers and pipeline frameworks reduce the burden of repetitive tasks and minimize the risk of inconsistent settings across files.

The [nf-core documentation](https://nf-co.re/docs) describes community pipeline standards that are particularly relevant for large scale projects. These pipelines provide preconfigured processing steps with documented parameters, and they enforce consistency across all input files. Researchers should assess whether an existing community pipeline matches their experimental design before building a custom workflow from individual tools.

The [Galaxy Training Network](https://training.galaxyproject.org/) offers another option for projects of intermediate scale. The web based platform allows researchers to build and share workflows without command line expertise, and the recorded parameter history supports reproducibility. This approach suits laboratories that process projects intermittently instead of maintaining a dedicated bioinformatics infrastructure.

### Sample Complexity and Purity Considerations

Sample complexity directly influences preprocessing decisions. A purified protein sample or a simple mixture requires less stringent quality control than a whole cell lysate or a biofluid sample. The [viewpoint article on plasma extracellular vesicles](https://pubmed.ncbi.nlm.nih.gov/40955644) highlights how biofluid samples contain many components with physicochemical properties similar to the target particles, creating challenges for data interpretation. For such complex samples, the preprocessing pipeline must include additional quality checks and may require more conservative identification thresholds.

For samples with expected contamination, such as common laboratory contaminants or host proteins in pathogen studies, the database selection step becomes more involved. The framework should include a contaminant database and a plan for evaluating the species distribution of identifications. A sample that yields a high proportion of contaminant identifications may indicate a sample preparation problem that preprocessing cannot correct.

### Instrument Type and Data Format Considerations

The instrument platform determines the raw file format and influences the conversion and calibration steps. Researchers should document the instrument model and acquisition settings before choosing conversion parameters. The [SpectiCal publication](https://pubmed.ncbi.nlm.nih.gov/38538550) demonstrates that calibration errors are common across instruments and that per file calibration based on known low mass ions improves identification quality. The decision to apply calibration should be based on evidence from initial searches instead of applied universally.

The conversion step requires attention to the specific vendor format. Thermo .raw files, Bruker .d folders, and SCIEX .wiff files each have distinct conversion considerations. The [EMBL-EBI training portal](https://www.ebi.ac.uk/training) provides educational material on data standards that helps researchers understand the relationship between vendor formats and the open mzML standard before beginning conversion.

### Downstream Analysis Goal Assessment

The intended downstream analysis should inform preprocessing choices from the start. A project focused solely on peptide identification can use a simpler pipeline than one that requires label free quantification or isobaric labeling quantification. The [MS-DAP study](https://pubmed.ncbi.nlm.nih.gov/36541440) demonstrated that the choice of normalization and statistical analysis methods has major effects on downstream results, and these choices depend on the quality and consistency of the preprocessing steps.

For label free quantification, the preprocessing pipeline must preserve intensity information accurately. Centroiding parameters affect peak area calculations, and inconsistent peak picking across files introduces quantitative noise. For isobaric labeling, the extraction of reporter ions from the low mass region requires accurate peak picking in that region, and the [SpectiCal tool](https://pubmed.ncbi.nlm.nih.gov/38538550) can identify artifacts that might interfere with reporter ion quantification.

### Decision Matrix for Common Scenarios

The following decision matrix summarizes the recommended preprocessing approach for common experimental scenarios. This matrix serves as a starting point, and researchers should adjust based on their specific samples and instruments.

| Scenario | Recommended Approach | Key Considerations |
| --- | --- | --- |
| Small pilot study, purified protein | Manual processing with individual tools | Close inspection of each step, minimal automation needed |
| Large cohort study, whole cell lysate | Community pipeline or workflow manager | Consistent settings across files, automated quality reporting |
| Biofluid sample with low abundance targets | Extended quality control, per file calibration | Contaminant handling, conservative identification thresholds |
| Cross linking mass spectrometry | Specialized search pipeline | Cross link aware search engine, dedicated visualization tools |
| Peptidomics without digestion | Adjusted search parameters | No enzyme specificity, small peptide mass range |

### Parameter Selection Based on Instrument Mass Accuracy

The precursor and fragment mass tolerance settings should reflect the instrument mass accuracy and the calibration status of the data. A high resolution instrument with routine calibration can use tight tolerances, while a lower resolution instrument or data with known calibration drift requires wider tolerances. The framework recommends starting with the instrument manufacturer specifications and then adjusting based on the observed mass error distribution from an initial search.

The [NCBI data resources](https://www.ncbi.nlm.nih.gov/) provide information about sequence databases that is relevant to parameter selection. The database size affects the search space and therefore the statistical significance of identifications. A larger database requires more stringent score thresholds to maintain the same false discovery rate.

### Implementation Steps for the Decision Framework

Step one is to document the project characteristics. Record the number of files, the instrument platform, the sample type, the expected protein complexity, and the downstream analysis goal. This documentation forms the basis for all subsequent decisions.

Step two is to select the processing approach based on project scale. Choose between manual processing, a workflow manager, or a community pipeline. The [nf-core documentation](https://nf-co.re/docs) and the [Galaxy Training Network](https://training.galaxyproject.org/) provide resources for the latter two options.

Step three is to determine the quality control intensity based on sample complexity. Biofluid samples and samples with expected contamination require more extensive quality checks than purified samples. The [viewpoint article on plasma extracellular vesicles](https://pubmed.ncbi.nlm.nih.gov/40955644) provides context for the challenges of complex biofluid samples.

Step four is to set initial search parameters based on instrument specifications and then refine them based on the observed mass error distribution. Record all parameter changes and the rationale for each change.

Step five is to evaluate the results against the decision criteria. If the identification numbers are unexpectedly low or the mass error distribution is broad, revisit the preprocessing decisions instead of forcing the data through the existing pipeline.

### Common Failure Patterns in Tool Selection

The first failure pattern is using a pipeline designed for a different instrument platform. Conversion settings and peak picking parameters that work well for one vendor format may produce poor results for another. Researchers should verify that their chosen tools support their specific instrument format.

The second failure pattern is applying a complex pipeline to a simple project. The overhead of workflow management and containerization may not be justified for a small pilot study, and the added complexity can obscure the preprocessing steps.

The third failure pattern is ignoring the downstream analysis requirements during preprocessing. A pipeline optimized for identification may not preserve the quantitative information needed for differential expression analysis, requiring the preprocessing to be redone.

The fourth failure pattern is failing to document the decision rationale. When the preprocessing choices are not recorded with their justification, the analysis cannot be audited or reproduced. The [Carpentries lessons](https://carpentries.org/lessons) cover foundational skills for reproducible research, including documentation practices.

### Records and Measurements for the Decision Framework

The decision framework requires its own record keeping. For each project, record the project characteristics that informed the preprocessing choices, the tools selected, the parameters used, and the rationale for each decision. This record should also include the results of any preliminary searches that informed parameter refinement.

The [Bioconductor project](https://bioconductor.org/) provides R packages that can generate standardized reports documenting preprocessing steps and quality metrics. These reports serve as a record of the decisions made and the evidence supporting them.

### Professional Escalation Criteria for Tool Selection

Researchers should escalate to more experienced colleagues when the decision framework does not produce satisfactory results. If the initial search yields very few identifications despite following the recommended approach, the problem may be in the tool selection or parameter configuration. An experienced practitioner can review the preprocessing steps and identify the issue.

If the researcher is uncertain about the appropriate pipeline for an unfamiliar sample type or instrument platform, consultation with a core facility is appropriate. Core facility staff have experience with a range of instruments and sample types and can provide guidance on tool selection and parameter configuration.

If the downstream analysis reveals inconsistencies that cannot be traced to the biological question, the preprocessing decisions should be audited. The records described above make this audit possible, and the [nf-core documentation](https://nf-co.re/docs) provides standards for workflow documentation that support such audits.

## Frequently Asked Questions

### What is the difference between profile data and centroided data in mass spectrometry?

Profile data represent each peak as a continuous series of data points across the mass to charge range, preserving the full peak shape. Centroided data reduce each peak to a single data point representing the monoisotopic mass and intensity. Database search engines expect centroided data because they compare observed peaks to theoretical fragment ions. The conversion from profile to centroided data is called peak picking or centroiding, and it must be performed consistently across all files in a project.

### Why is mzML conversion necessary if my instrument software can export peak lists?

Vendor software can export peak lists, but the formats are often incomplete or incompatible with downstream tools. The mzML format is an open standard developed by the Proteomics Standards Initiative that represents spectra and metadata in a vendor neutral way. Converting to mzML ensures that any analysis tool can read the data and that the full spectral information is preserved. The [EMBL-EBI training portal](https://www.ebi.ac.uk/training) provides educational material on data standards in bioinformatics.

### How do I choose the right protein database for my search?

The database should contain the protein sequences expected in the sample. For a well characterized organism, the reference proteome from a major database such as [NCBI](https://www.ncbi.nlm.nih.gov/) is usually appropriate. For samples that may contain contaminants, add a common contaminants database. For less characterized organisms, consider a broader database that includes related species. Record the database name, version, and download date for reproducibility.

### What is the target decoy strategy and why is it important?

The target decoy strategy adds reversed or shuffled protein sequences to the search database. Matches to decoy sequences estimate the number of false matches to target sequences. The false discovery rate is calculated as the ratio of decoy hits to target hits, and a threshold such as 1 percent is applied to filter the results. This strategy is important because search scores alone do not distinguish true from false matches.

### When should I apply mass calibration to my data?

Apply mass calibration when the search results show systematic mass errors, such as a consistent offset across the m/z range or a drift with retention time. The [SpectiCal tool](https://pubmed.ncbi.nlm.nih.gov/38538550) computes per file calibrations based on known low mass ions and generates calibrated mzML files. Calibration is preferable to widening the search tolerance because widening tolerance increases the false discovery rate.

### What quality control checks should I perform before database searching?

Check the total ion chromatogram and base peak chromatogram for each file to identify runs with no peaks or unusual baselines. Check the number of MS2 spectra per file and the precursor charge state distribution. After conversion, verify that the mzML files are readable and contain the expected number of spectra. The [Galaxy Training Network](https://training.galaxyproject.org/) provides tutorials that include quality control steps for proteomics data.

### How do I handle proteins identified by a single peptide?

Proteins identified by a single peptide are less confident than proteins identified by multiple peptides. Some studies require a minimum of two peptides per protein, while others accept single peptide identifications with additional evidence such as a high search score and a consistent mass error. The decision should be made before the analysis and recorded in the methods. The protein inference step should report the number of single peptide identifications so that readers can assess their contribution.

### What records should I keep for reproducible preprocessing?

Keep the raw file inventory with acquisition dates and sample annotations, the conversion log with tool versions and settings, the search parameter file with database version and all search settings, and the quality control log with metrics at each checkpoint. These records allow the analysis to be audited and reproduced. The [nf-core documentation](https://nf-co.re/docs) describes community standards for reproducible workflows that apply to proteomics preprocessing.

## Related Bioinformatics Guides

- [Proteomics Mass Spectrometry: From Sample Preparation to Data Analysis](/knowledge/bioinformatics/proteomics-mass-spectrometry-from-sample-preparation-to-data-analysis)
- [Mass Spectrometry-Based Proteomics: Data Analysis Pipelines and Tools](/knowledge/bioinformatics/mass-spectrometry-based-proteomics-data-analysis-pipelines-and-tools)
- [Genomic Data Processing: From Raw Sequencing to Analysis-Ready Files](/knowledge/bioinformatics/genomic-data-processing-from-raw-sequencing-to-analysis-ready-files)
- [Proteomics Data Analysis Workflow: From Raw Spectra to Biological Insights](/knowledge/bioinformatics/proteomics-data-analysis-workflow-from-raw-spectra-to-biological-insights)
- [Single-Cell Sequencing Analysis Pipeline: From Raw Data to Biological Insights](/knowledge/bioinformatics/single-cell-sequencing-analysis-pipeline-from-raw-data-to-biological-insights)

## 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.
- [MS-DAP Platform for Downstream Data Analysis of Label-Free Proteomics Uncovers Optimal Workflows in Benchmark Data Sets and Increased Sensitivity in Analysis of Alzheimer's Biomarker Data.](https://pubmed.ncbi.nlm.nih.gov/36541440). Journal of proteome research, 2023.
- [Challenges and Opportunities in State-of-the-Art Proteomics Analysis for Biomarker Development From Plasma Extracellular Vesicles.](https://pubmed.ncbi.nlm.nih.gov/40955644). Proteomics, 2026.
- [Cross-ID: Analysis and Visualization of Complex XL-MS-Driven Protein Interaction Networks.](https://pubmed.ncbi.nlm.nih.gov/30575379). Journal of proteome research, 2019.
- [SpectiCal: m/z Calibration of MS2 Peptide Spectra Using Known Low Mass Ions.](https://pubmed.ncbi.nlm.nih.gov/38538550). Journal of proteome research, 2024.
- [Peptidomics of Neuropeptidergic Tissues of the Tsetse Fly Glossina morsitans morsitans.](https://pubmed.ncbi.nlm.nih.gov/26463237). Journal of the American Society for Mass Spectrometry, 2015.

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