[A new peptide retention time prediction method for mass spectrometry based proteomic analysis by a serial and...

Area of Science:

• Proteomics
• Analytical Chemistry
• Computational Biology

Context:

• Online reversed-phase liquid chromatography (RPLC) coupled with mass spectrometry is crucial for large-scale protein identification.
• Accurate peptide retention time (RT) prediction is vital for distinguishing true positive from false positive identifications.
• Existing sequence-based RT prediction methods suffer from low accuracy due to nonlinear mobile phase gradients and inter-peptide interactions.

Purpose:

• To develop a novel method, serial and parallel support vector machine (SP-SVM), to accurately predict peptide retention times.
• To address the limitations of existing methods by characterizing nonlinear organic phase concentration effects and peptide interactions.
• To improve the reliability of peptide identification in proteomics.

Summary:

• The proposed SP-SVM method utilizes multiple support vector machine (SVM) models, including a support vector regression (SVR) for training, and specific SVMs (C-SVM, 1-SVR, s-SVR) to predict peptide RTs.
• A further SVM model (n-SVR) is employed for normalizing RTs to account for peptide interactions during chromatographic separation.
• This integrated approach significantly enhances prediction accuracy, achieving a coefficient of determination of 0.95 and minimizing prediction errors.

Impact:

• The SP-SVM method achieves state-of-the-art performance in peptide RT prediction, with over 95% of predictions having <20% error and over 70% having <10% error.
• Provides a robust framework for incorporating peptide interactions into chromatographic separation analysis.
• Offers a significant advancement in validating peptide identifications from mass spectrometry data, improving the overall quality of proteomics studies.