Machine learning application to predict binding affinity between peptide containing non-canonical amino acids and...

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Class 1 major histocompatibility complexes (MHC-I) present peptides to CD8+ T cells, crucial for immune responses.
• Peptide binding affinity to MHC-I influences immunogenicity, making prediction valuable for identifying potential antigens.
• Existing MHC-I binding predictors lack support for non-canonical amino acids (NCAAs) and post-translational modifications, limiting their application.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting MHC-I binding affinity of epitopes containing NCAAs.
• To compare the model's performance against established regression methods.
• To provide a computational tool for designing and optimizing peptides with NCAAs for therapeutic development.

Main Methods:

• Development of a machine learning application to quantify binding affinity.
• Inclusion of antigens with explicitly labeled post-translational modifications and NCAAs.
• Performance evaluation using 5-fold cross-validation, R-squared, and RMSE metrics.

Main Results:

• The proposed machine learning model demonstrated robust performance in predicting MHC-I binding affinity for peptides containing NCAAs.
• Achieved an R-squared value of 0.477 and a root-mean-square error (RMSE) of 0.735 via 5-fold cross-validation.
• Indicated strong predictive capability for peptides incorporating NCAAs.

Conclusions:

• The developed model offers a valuable tool for computational design and optimization of peptides with NCAAs.
• Facilitates the acceleration of novel peptide-based therapeutics with enhanced properties and efficacy.
• Addresses a critical gap in current MHC-I binding prediction tools by incorporating NCAAs.