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

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Class I major histocompatibility complexes (MHC-I) present peptides on nucleated cells to activate CD8+ T cells.
• Peptide immunogenicity correlates with binding affinity to the MHC-I binding groove.
• Existing prediction tools lack the ability to handle non-canonical amino acids (NCAAs) or post-translational modifications.

Purpose of the Study:

• To develop and evaluate a machine learning application for predicting MHC-I binding affinity of peptides containing NCAAs.
• To compare the performance of the proposed model against commonly used regression methods.
• To provide a tool for computational design and optimization of NCAAs-containing peptides.

Main Methods:

• Development of a machine learning model to quantify binding affinity.
• Inclusion of explicitly labeled post-translational modifications and NCAAs in the prediction.
• Performance evaluation using 5-fold cross-validation, R2, and RMSE metrics.

Main Results:

• The proposed machine learning model demonstrated robust performance.
• A 5-fold cross-validation yielded an R2 value of 0.477 and an RMSE of 0.735.
• The model shows strong predictive capability for peptides containing NCAAs.

Conclusions:

• The developed machine learning application accurately predicts MHC-I binding affinity for peptides with NCAAs.
• This tool is valuable for the computational design and optimization of peptides incorporating NCAAs.
• The findings can accelerate the development of novel peptide-based therapeutics with enhanced properties.