GeoPPI

GeoPPI predicts changes in protein-protein binding affinity caused by amino acid mutations using deep learning and geometric representations to inform protein engineering and drug design research.

Key Features:

• Structure-Based Deep Learning: Uses three-dimensional protein structural data and a self-supervised learning scheme to derive geometric representations that capture topological features of protein structures.
• Gradient-Boosting Trees for Prediction: Feeds the learned geometric representations into gradient-boosting trees to predict changes in binding affinity caused by mutations.
• Handling Single and Multi-point Mutations: Predicts effects of both single-point and multi-point amino acid substitutions on protein-protein interaction affinity across benchmark datasets.
• State-of-the-Art Performance: Demonstrated superior predictive performance compared to existing methods through extensive experimental validation.
• Application to SARS-CoV-2 Research: Applied to estimate binding affinity differences between antibodies targeting the receptor-binding domain (RBD) of the SARS-CoV-2 S protein.

Scientific Applications:

• Protein Engineering: Guides design of proteins with modified interaction affinities by predicting mutation impacts on protein-protein binding.
• Drug Design: Supports development of therapeutics by predicting how mutations alter therapeutic protein–protein interactions.
• Viral Mechanism and Antibody Analysis: Enables analysis of viral-host and antibody-antigen interaction changes, exemplified by studies on antibodies binding the SARS-CoV-2 RBD.

Methodology:

GeoPPI learns geometric representations from three-dimensional protein structures using self-supervised learning to capture topological features, then uses these representations as input features for gradient-boosting trees trained to predict changes in protein-protein binding affinity resulting from amino acid mutations.