POPI

POPI predicts the immunogenicity of peptides that bind to Major Histocompatibility Complex (MHC) class I molecules using support vector machine (SVM) models trained on informative physicochemical properties.

Key Features:

• Physicochemical Property Mining: Mines a feature set of physicochemical properties from MHC class I binding peptides to identify properties correlated with peptide immunogenicity.
• Support Vector Machine (SVM) Framework: Implements an SVM-based classification system for predicting peptide immunogenicity.
• Bi-Objective Genetic Algorithm: Uses an inheritable bi-objective genetic algorithm to select m = 23 physicochemical properties from a pool of 531 candidates and to optimize SVM parameters.
• Performance Evaluation: Validated on 428 human MHC class I binding peptides categorized into four immunogenicity classes using leave-one-out cross-validation, achieving 64.72% accuracy and outperforming ALIGN (54.91%) and PSI-BLAST (53.23%).
• Dataset (PEPMHCI): Development and validation used the PEPMHCI dataset of MHC class I binding peptides.

Scientific Applications:

• Vaccine Design: Predicts peptides likely to elicit immune responses to inform rational peptide-based vaccine candidate selection.
• Immunoinformatics Research: Facilitates analysis of factors influencing peptide immunogenicity and studies of the antigen-processing pathway.

Methodology:

Physicochemical properties were mined from MHC class I binding peptides; an SVM classifier was trained with parameters tuned via an inheritable bi-objective genetic algorithm that selected m = 23 properties from 531 candidates; performance was assessed by leave-one-out cross-validation on 428 peptides and compared with ALIGN and PSI-BLAST.