Deep geometric representations for modeling effects of mutations on protein-protein binding affinity

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Accurately modeling the impact of amino acid mutations on protein-protein interactions is vital for protein engineering and drug design.
• Predicting changes in binding affinity due to mutations is a challenging but critical task.

Purpose of the Study:

• To develop GeoPPI, a novel structure-based deep-learning framework for predicting binding affinity changes upon mutations.
• To leverage geometric representations of protein structures for accurate prediction of mutation effects.

Main Methods:

• GeoPPI utilizes a self-supervised learning scheme to derive geometric representations encoding protein structure topology.
• These learned representations serve as features for gradient-boosting trees to predict binding affinity alterations.
• The framework is based on the three-dimensional structure of proteins.

Main Results:

• GeoPPI successfully learns meaningful features characterizing atomic interactions within protein structures.
• The framework achieves state-of-the-art performance in predicting binding affinity changes for single and multi-point mutations across six benchmark datasets.
• GeoPPI accurately estimates binding affinity differences for SARS-CoV-2 antibodies and the S protein receptor-binding domain (RBD).

Conclusions:

• GeoPPI demonstrates significant potential as a powerful computational tool for protein design and engineering.
• The framework provides accurate predictions for mutation-induced binding affinity changes, aiding in rational protein design.
• The study highlights the utility of deep learning in understanding and predicting protein-protein interactions.