Unifying protein inference and peptide identification with feedback to update consistency between peptides

Area of Science:

• Proteomics
• Mass Spectrometry Data Analysis
• Bioinformatics

Background:

• Accurate peptide identification and protein inference are crucial for mass spectrometry-based proteomics.
• Existing methods often face challenges in distinguishing true positives from false positives and maximizing protein coverage.

Purpose of the Study:

• To develop an integrated method for unifying peptide identification and protein inference.
• To enhance the accuracy and discrimination power of peptide scoring.
• To improve the coverage and reduce false positives in protein inference.

Main Methods:

• A novel method for processing peptide identification reports from database search engines.
• Integration of protein inference feedback to update peptide-to-peptide adjacency matrices.
• Regularization of peptide scores using the adjacency matrix and logistic regression (LR) for probability computation.
• Calculation of protein scores based on peptide probabilities and selection of multiple peptide matches per MS/MS spectrum.

Main Results:

• The proposed method robustly assigns accurate probabilities to identified peptides.
• Demonstrated higher discrimination power compared to PeptideProphet in distinguishing correct from incorrect peptide identifications.
• Achieved superior performance over ProteinProphet by inferring more true positive proteins and fewer false positive proteins at equivalent false positive rates.
• Significantly increased the coverage of inferred proteins.

Conclusions:

• The developed integrated approach offers a robust and accurate solution for peptide and protein identification in mass spectrometry.
• The feedback mechanism from protein inference to peptide identification enhances scoring accuracy and discrimination.
• This method leads to improved protein inference, increased coverage, and better overall performance in proteomics studies.