Dynamics-Based Peptide-MHC Binding Optimization by a Convolutional Variational Autoencoder: A Use-Case Model for...

Area of Science:

• Computational biology
• Immunoinformatics
• Machine learning

Background:

• Selecting optimal antigens is crucial for developing effective antigen-specific immunotherapies.
• Understanding antigen-major histocompatibility complex (MHC) binding is essential for this selection process.

Purpose of the Study:

• To apply a combined machine learning-molecular dynamics (ML-MD) approach, named CASTELO, to identify per-residue antigen binding contributions.
• To design novel antigens with enhanced MHC-II binding affinity for a type 1 diabetes-related system.

Main Methods:

• Trained a convolutional variational autoencoder (CVAE) on molecular dynamics (MD) trajectories from 48 systems across four antigens and four HLA serotypes.
• Developed new machine learning metrics, including a structure-based anchor residue classification model and cluster comparison scores.
• Utilized a small-molecule lead optimization algorithm as a foundation.

Main Results:

• ML-MD predictions showed strong agreement with experimental binding data and free energy perturbation-predicted binding affinities.
• Developed ML-MD metrics demonstrated independence from traditional MD stability metrics like contact area and root-mean-square fluctuations (RMSF).
• Identified per-residue antigen binding contributions and designed novel antigens with increased MHC-II binding affinity.

Conclusions:

• The CASTELO ML-MD approach effectively identifies antigen binding contributions and facilitates the design of novel antigens.
• Structure-based deep learning techniques show significant promise for advancing antigen-specific immunotherapy design.
• The developed ML-MD metrics offer a more accurate reflection of binding affinity compared to traditional MD stability metrics.