Navigating the Mass Spectrometry-Based Proteomic Data Using Free Computational Tools

Summary

Mass spectrometry-based proteomic data is available in open databases and accessible using free tools. Given the complexity of database searches and descriptions, many biologists lack the knowledge to utilize these datasets. Here, we provide a guide on using free tools for basic proteomic data searches.

Abstract

Mass spectrometry (MS)-based proteomics data, including Data-Dependent Acquisition (DDA) and Data-Independent Acquisition (DIA), are widely used in biological research. However, the application of these datasets in validation studies is still limited due to the lack of clear demonstrations on how to effectively search and analyze proteomic data. To fill this gap, we selected one DDA and one DIA dataset deposited in the PRoteomics IDEntifications Database (PRIDE) data repository to better illustrate the proteomic data analysis workflow and downstream post-processing of protein search results. For demonstration purposes, we used two free computational tools: FragPipe (v22.0) for DDA datasets and DIA-NN (2.1.0) for DIA datasets. Post-processing steps, such as generating volcano plots and lists of dysregulated proteins, were demonstrated using R code. This study provides basic protocols for searching and analyzing proteomic data, serving as an essential beginner's guide to effectively handle proteomic datasets. Through this work, we aim to empower researchers with the knowledge necessary to leverage proteomic data in their biological investigations.

Introduction

The field of proteomics has been revolutionized by the widespread adoption of open-access data repositories and free software tools¹, which have been instrumental in advancing research capabilities and promoting a culture of data sharing. The establishment of centralized resources, such as PRIDE², and the development of standardized data formats by the Human Proteome Organization (HUPO) Proteomics Standards Initiative (PSI)³ have created an unprecedented opportunity for data reuse and integration. However, the analysis and interpretation of proteomic data still present significant technical barriers, particularly for biologists and physician scientists who wish to perform orthogonal validation of their own experimental results⁴. This manuscript directly addresses this challenge by presenting a vendor-agnostic computational workflow to provide an accessible framework for researchers to independently query, analyze, and integrate public proteomics data with their own findings, thereby facilitating robust orthogonal validation without the need for additional wet-lab experiments. This work is particularly appropriate for researchers seeking to orthogonally validate the differential expression of specific proteins identified in their own experiments and generate novel hypotheses by exploring the expression patterns of their proteins of interest across a wide range of published studies. By providing a step-by-step guide, we aim to empower a wider range of scientists to harness the full potential of public proteomics data, ultimately enhancing the robustness and impact of their research.

In mass spectrometry (MS)-based proteomics, MS1 and MS2 scans are fundamental to data acquisition. MS1 scans, or survey scans, detect and measure the mass-to-charge (m/z) ratios of all ions in a sample, providing a comprehensive overview of the ion population⁵. MS2 scans, also called fragmentation scans, involve isolating and fragmenting specific precursor ions chosen from the MS1 scan, generating fragment ion spectra for peptide identification⁶.

Data-dependent acquisition (DDA) selects the most abundant ions from MS1 scans for MS2 fragmentation, producing simpler spectra that streamline analysis. However, DDA can lead to stochastic sampling and missing values in complex samples, limiting reproducibility⁷. In contrast, data-independent acquisition (DIA) systematically fragments all ions within predefined m/z windows, ensuring comprehensive coverage and minimizing missing values. This enhances quantitative accuracy and reproducibility⁸, though DIA demands advanced computational tools to deconvolve its highly complex MS2 spectra⁹.

PRIDE (https://www.ebi.ac.uk/pride)²i s a leading repository within the ProteomeXchange Consortium (https://www.proteomexchange.org)¹⁰, a global platform for sharing proteomics data. Users can efficiently navigate this vast resource using keyword searches to locate datasets of interest.

DIA-NN¹¹, a software suite leveraging deep neural networks, is a gold-standard tool for analyzing data-independent acquisition (DIA) proteomics data. It excels in identification and quantification accuracy, making it particularly effective for processing complex DIA datasets¹². For data-dependent acquisition (DDA) workflows, MSFragger¹³, integrated into the FragPipe pipeline, offers ultrafast peptide identification. Its innovative fragment-ion indexing approach enables rapid spectral matching, outperforming traditional tools by over 100-fold in speed.

Equipped with these comprehensive databases and advanced tools, researchers can readily reuse proteomic data for cross-study validation and novel biological discoveries. This accessibility has driven the identification of numerous protein biomarkers across diverse diseases^14,15, improved prognosis prediction¹⁶, and enabled molecular subtype classification based on proteomic profiles¹⁷.

Protocol

NOTE: See Figure 1for an overview of the workflow and Table of Materials for detailed descriptions of the software used.

Figure 1: Overview of study design. Please click here to view a larger version of this figure.

1. Example file and software setup

NOTE: All applications are free for individual research use and can be run on a personal computer (Windows and Linux are supported).

1. Create a new directory for a project.
2. Download JMU_LM2_8526_55.raw, JMU_LM2_8568_51.raw, JMU_LM2_8696_37.raw, JMU_LM2_2112_87.raw, JMU_LM2_2195_91.raw, JMU_LM2_2377_73.raw, 20150127_liver_XH03_T_01.raw, 20150127_liver_XH03_T_02.raw, 20150127_liver_XH03_T_03.raw, 20150127_liver_XH03_P_01.raw, 20150127_liver_XH03_P_02.raw, 20150127_liver_XH03_P_03.raw from PRIDE.
   NOTE: The two groups of files can be downloaded from https://www.ebi.ac.uk/pride/archive/projects/PXD039273 and https://www.ebi.ac.uk/pride/archive/projects/PXD006512 separately (Figure 2).
3. Download and install the required software tools.
   1. Download the latest version of DIA-NN as a .msi file. DIA-NN is freely available as a .msi file for Windows or .zip file for Linux from https://github.com/vdemichev/DiaNN/releases/tag/2.0
   2. Download the latest version of MSFileReader as an .exe file. MSFileReader is freely available as an .exe file from https://github.com/thermofisherlsms/MSFileReader/blob/main/MSFileReader_x64.exe, which is required to process raw formatted Thermo Fisher DIA data.
   3. Download the latest version of .NET SDK as an .exe file. .NET SDK is freely available as an .exe file from https://dotnet.microsoft.com/en-us/download, which is required in Windows to execute MSFileReader.
   4. Download the latest version of FragPipe as a .zip file. FragPipe is freely available as a .zip file from https://github.com/Nesvilab/FragPipe/releases. The zip file with jre is recommended as it contains a Java runtime for Windows, which is needed to execute FragPipe. MSFragger is embedded in the FragPipe.
4. Download the latest version of R and Rstudio and R script files from Supplementary File 1 and Supplementary File 2. R and RStudio can be downloaded freely from https://cran.rstudio.com/bin/windows/base/R-4.5.0-win.exe and https://posit.co/download/rstudio-desktop/.
5. Unzip the compressed files and click on the .exe or .msi Files to install the applications.

2. DIA data analysis using DIA-NN

1. Download the reference proteome fasta file from UniProt. To download the Homo sapiens reference proteome (isoform information not included) go to https://www.uniprot.org/proteomes/UP000005640 (Figure 3A).
2. Click on the Add FASTA button to load the reference proteome file into DIA-NN.
3. Select the two options FASTA digest for library-free search /library generation and Deep learning-based spectra, RTs and lMs prediction under Precursor ion generation part. Click on the Run button to generate a predicted spectral library (Figure 3B).
   1. Once the predicted spectral library file was generated, reuse that for other experiments within the same experimental acquisition settings through clicking on Spectral button to add the predicted library.
4. Unselect the two options under the Precursor ion generation part and click on the Type button that correlates with the file format under the Input part to load the DIA data.
5. Set the Mass Accuracy and MS1 Accuracy under Algorithm to 0.0 ppm.
6. Adjust the precursor and fragment mass range settings under Precursor ion generation part based on the experimental settings.
7. Keep the other default settings.
   1. DIA-NN's precursor generation uses Trypsin/P (cleaves after K/R unless followed by proline) with 1 missed cleavage to model partial digestion; disable N-term M excision for non-canonical peptides (e.g., immunopeptidomics) or increase missed cleavages to 2 for longer peptides.
   2. Explicitly enable Variable modifications (disabled by default) for post-translational modification (PTM) studies (e.g., Ox (M) for oxidation, K-GG for ubiquitination), and expand peptide filters (length: 7-30 aa; charge: 1-4+; precursor m/z: 300-1800) according to the experimental settings (e.g., length:5-50 aa, charge=1-6) for large PTMs or non-tryptic digests.
   3. Algorithm settings leverage auto-calibration (mass accuracy=0.0 ppm) for precise mass alignment—override with manual tolerances (e.g., 10 ppm) for low-resolution instruments—and Match Between Runs (MBR) transfers IDs across runs (disable for heterogeneous samples).
   4. Neural network scoring boosts identification confidence (retain unless poor data quality), while Quant Unlabeled Mass Spectrometry (UMS) enables high-precision label-free quantification; switch to MS2 quantification for isobaric tags (e.g., TMT). Use library-free search only with high-resolution DIA data to bypass spectral library requirements.
8. Click on the Run button. Wait for results until Finished is shown in the operation interface indicating the run is finished (Figure 3C).

3. DDA data analysis using FragPipe

1. Click on Fragpipe icon in the ~\FragPipe\fragpipe\bin folder after installation. All the dependent settings can be seen in the Config tab. Check whether the three modules (MSFragger, IonQuant, diaTracer, DIA-NN and Python) exist on your computer. If not, download them by clicking on Download/Update or Download (Figure 4A).
2. Switch to next tab, Workflow. Select Default for the workflow and click on Load workflow. Click on Add files to add file paths. Assign experiment name and bioreplicate number under Assign files or just leave it (Figure 4B).
3. Now, click on Database and turn to that tab. Load a fasta file or download any fasta files according to the species of the samples. While downloading, choose Reviewed sequences only, Add decoys and Add common contaminants options for a simple run (Figure 4C).
4. Then click on MSFragger to change the tab. Choose the Closed Search default config and click on Load.
   NOTE: The closed search is basically a search for the modifications that you have set while the open search will explore all the possible modifications from the data, which is more time-and memory-consuming.
5. For Peak matching, keep the default settings except that choosing None for Calibration and Optimization is recommended, as it will save time.
6. For protein digestion, adjust the parameters according to the experimental conditions. Keep the following default settings (Figure 5A).
   NOTE: The precursor and fragment mass tolerances (± 20 ppm) define the maximum allowed mass error for matching experimental spectra to theoretical peptides, with tighter values recommended for high-resolution instruments. Enzyme rules (strict trypsin) specify cleavage specificity (after K/R, unless followed by P), with missed cleavages (2) permitting partially digested peptides. Peptide length (5-50 residues) and mass (500-5000 Da) filters exclude unlikely candidates. Variable modifications (e.g., +15.9949 Da on methionine for oxidation, +79.96633 Da on S/T/Y for phosphorylation) allow detection of chemical alterations, while Max combinations (5) limit simultaneous modifications per peptide to manage search space. Max variable mods per site (e.g., 3 on M) controls modification frequency. Split database improves efficiency for large proteomes, and isotope error (0/1/2) accounts for misassigned isotopic peaks. These settings collectively balance sensitivity, specificity, and computational load in peptide identification.
7. Change to Validation tab. Unselect Predict RT and Predict spectra, which is for DIA data (Figure 5B).
8. Click on Quant (MS1) tab. Select Run MS1 quant and click on Load Quant defaults. Select IonQuant and keep the default settings (Figure 6A). Disable MBR for unstable retention times or disable Intensity normalization if global scaling biases result; enable Keep index on disk to conserve RAM with large datasets.
   NOTE: Quant (MS1) defaults to optimize label-free quantification using MaxLPQ (MaxLFQ-based Protein Quantification, an algorithm extending the MaxLFQ method to protein groups), requiring ≥1 unique ion per protein (increase to ≥2 for higher stringency at the cost of coverage). Match Between Runs (MBR) transfers identifications across runs using retention time (RT) tolerance (default 0.4 min; tighten to 0.2 min for stable chromatography or loosen for gradient variability) and ion mobility (IM) tolerance (default 0.05 1/k0 [inverse reduced mobility]; adjust based on IM precision), filtered by ≤1% FDR (False Discovery Rate; default MBR ion FDR=0.01; lower to 0.005 for rigor). Protein inference uses uniquer+rascor (razor peptide assignment for protein grouping). Feature detection employs 10 ppm mass tolerance (increase to 20 ppm for low-resolution MS or decrease to 5 ppm for high-resolution). 
9. Now, click on the Run tab. Select Output directory. Click on RUN to start analyzing the data (Figure 6B).
   NOTE: For label-free proteomics data not focusing on modifications or using label-based quantification, PTMs, Glyco, and Quant (Isobaric) tabs are not used.

4. DIA-NN/FragPipe downstream analysis

NOTE: Detailed codes used in this study are included as Supplementary File 1 and Supplementary File 2. The used package versions in this study can be found in Supplementary File 3.

1. Open RStudio and read the protein expression data generated by DIA-NN or FragPipe. Protein group matrix data ending with .pg_matrix.tsv from DIA-NN and combined protein data ending with .tsv from FragPipe are normally used for analysis.
2. Remove the missing values without filling in NA and filter out the protein groups identified and quantified with less than two unique peptides.
   NOTE: The DIA figures in this study were generated by filtered protein groups, while the DDA figures were from unfiltered protein groups.
3. Normalize protein intensities using median scaling or quantile normalization across samples. DIA data in this study was normalized by median normalization while DDA data by quantile normalization to better demonstrate the two methods.
   NOTE: Median normalization is robust against outliers and suitable for scenarios where the majority of proteins remain unchanged across samples, making the median protein abundance a stable reference point. However, it may fail in scenarios with highly differentially expressed proteins¹⁸. Quantile normalization ensures a uniform distribution of protein abundances across samples, which is helpful when comparing samples with disparate distributions. It has been shown to reduce statistical power and may over-normalize data, potentially masking true biological differences¹⁸. In proteomics, median normalization is widely used in label-free studies to remove systematic biases related to MS instrument performance. Quantile normalization is preferred for large-scale datasets, differential expression analysis, and cluster analysis^19,20.
4. Calculate log2 fold changes (e.g., PDAC Serum versus Normal) and t-test p-values. Calculate corrected p-values using the Benjamini-Hochberg method. Define significance thresholds (e.g., |log2FC| > 1, p_adj (FDR) < 0.05).
   NOTE: The cutoff p-values utilized in this study's figures represent raw values, as employing adjusted p-values (p_adj) would eliminate all detectable proteins within the statistical power of these limited datasets.
5. Check the existence of missing/NA values and exclude them. Generate a volcano plot to visualize significant hits.
6. Select significantly dysregulated proteins. Apply Z-score normalization to protein expression values. Plot a clustered heatmap with sample group annotations.
7. Map proteins to Entrez IDs using org.Hs.eg.db. Perform GO enrichment (Biological Process) and KEGG pathway analysis.
   NOTE: Biological Process (BP) categories are commonly selected in Gene Ontology (GO) analysis because they describe the larger processes or biological programs accomplished by multiple molecular activities²⁰. These terms focus on cellular functions that go awry in diseases like cancer, including essential processes such as cell cycle regulation, metabolic processes, etc. In addition to BP, GO analysis also includes Molecular Function (MF) and Cellular Component (CC) categories. MF terms outline the biochemical activities of gene products, such as ion transport or DNA-binding capabilities. CC terms define the cellular locations where gene products carry out their functions, such as the nucleus, mitochondrial membrane, or cytoskeleton.
8. Query STRINGdb for interactions (confidence score ≥ 400). Generate an interactive network using visNetwork.

Results

To better illustrate the search for proteomic datasets, we selected two types of datasets for this analysis. Aiming to demonstrate dysregulated proteins in clinical contexts, we searched in PRIDE using keywords such as clinic. Specifically, we selected a dataset suitable for DIA analysis from the Clinical Proteomic Tumor Analysis Consortium (CPTAC)²¹, and another suitable for DDA analysis from the Chinese Human Proteome Project (CNHPP)²². Th...

Discussion

The presented protocol illustrates proteomic data analysis by integrating free tools like DIA-NN and FragPipe. Success hinges on critical steps such as accurate precursor ion selection and mass accuracy settings for spectral library generation in DIA-NN (step 2.3)²⁷, which are instrument-dependent and crucial for complete peptide identification. Similarly, meticulous database setup in FragPipe (step 3.4), including decoy generation and contaminant filtering, is vital for controlling FDR.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
.NET SDK v. 9.0.203	Microsoft		https://dotnet.microsoft.com/en-us/download
DIA-NN v. 2.1.0			https://github.com/vdemichev/DiaNN/releases/tag/2.0
FragPipe v. 22.0			https://github.com/Nesvilab/FragPipe/releases
MSFileReader v. 3.1	Thermo Fisher		https://github.com/thermofisherlsms/MSFileReader/blob/main/MSFileReader_x64.exe
R 4.5.0			https://cran.rstudio.com/bin/windows/base/R-4.5.0-win.exe
Rstudio v. 2024.12.1+563			https://posit.co/download/rstudio-desktop/