Semisupervised model-based validation of peptide identifications in mass spectrometry-based proteomics

Area of Science:

• Proteomics
• Computational Biology
• Biostatistics

Background:

• Accurate validation of peptide assignments to tandem mass spectrometry (MS/MS) spectra is crucial for reliable proteomic data analysis.
• Existing tools like PeptideProphet and decoy strategies offer methods for validation, but improvements in accuracy and robustness are needed, especially for challenging datasets.
• Integrating these approaches could enhance the reliability of peptide identification.

Purpose of the Study:

• To develop a robust statistical method for validating peptide assignments in MS/MS spectra.
• To combine the probabilistic modeling of PeptideProphet with the decoy strategy within a semisupervised framework.
• To improve the accuracy and robustness of computed probabilities for peptide identification.

Main Methods:

• Development of a semisupervised expectation-maximization (EM) algorithm.
• Construction of a Bayes classifier for peptide identification using a probability mixture model.
• Extension of PeptideProphet to incorporate decoy peptide matches.

Main Results:

• The combined approach demonstrated improved robustness and higher accuracy of computed probabilities.
• The method effectively controls the false discovery rate, even with complex datasets.
• Validation was successful across different database search programs (SEQUEST, MASCOT, TANDEM/k-score) and sample types (protein mixtures, human plasma).

Conclusions:

• The semisupervised EM algorithm integrating PeptideProphet and decoy strategies offers a more accurate and robust method for peptide identification validation.
• This approach enhances the control of false discovery rates in proteomics.
• The method is applicable to diverse and complex proteomic datasets, improving overall data reliability.