Performance Evaluation of MHC Class-I Binding Prediction Tools Based on an Experimentally Validated MHC-Peptide...

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Accurate prediction of peptide-MHC binding is vital for designing effective immunotherapies.
• Existing in silico peptide-MHC class-I binding prediction algorithms show variable performance based on algorithm, MHC type, and peptide length.

Purpose of the Study:

• To evaluate the performance of 13 peptide-MHC class-I binding prediction algorithms.
• To identify optimal prediction thresholds for increased sensitivity.
• To develop a tool for combining predictions from multiple algorithms.

Main Methods:

• Evaluated 13 algorithms using 8-11mer peptides from HPV16 E6/E7 proteins and prevalent HLA types.
• Experimentally verified in vitro HLA binding of synthesized peptides.
• Analyzed algorithm performance using ROC curves and calculated optimal decision thresholds.

Main Results:

• No single algorithm consistently outperformed others across all HLA types and peptide lengths.
• Commonly used thresholds showed low sensitivity (40%); optimized thresholds significantly improved prediction accuracy.
• Developed MHCcombine, a web application to integrate predictions from up to 13 algorithms.

Conclusions:

• Peptide-MHC binding prediction tool performance varies, necessitating careful selection and validation.
• Optimizing decision thresholds and combining multiple prediction tools can enhance the identification of potential immunotherapy epitopes.
• MHCcombine facilitates broader epitope discovery for immunotherapy development.