DDAffinity: predicting the changes in binding affinity of multiple point mutations using protein 3D structure.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning

Background:

• Mutations drive evolution by altering protein stability, interactions, structure, function, and expression.
• Existing computational methods struggle with predicting effects of multiple point mutations due to complex synergistic epistasis.

Purpose of the Study:

• To develop a novel computational method, DDAffinity, for predicting changes in protein binding affinity caused by multiple point mutations.
• To address limitations of existing methods by systematically accounting for local and global synergistic epistasis.

Main Methods:

• Developed DDAffinity, a spatial and sequential message passing neural network utilizing protein 3D structures.
• Employed k-nearest neighbor residue graphs for extracting protein pocket features.
• Utilized a two-step additive Gaussian noising strategy to capture global topological features.

Main Results:

• DDAffinity significantly improves prediction accuracy for multiple point mutations compared to state-of-the-art methods.
• Ablation studies confirm the effectiveness of DDAffinity's components.
• Demonstrated effectiveness in blind testing, SARS-CoV-2 variant analysis, and antibody optimization.

Conclusions:

• DDAffinity offers a robust and accurate approach for predicting the impact of multiple mutations on protein binding affinity.
• The method holds promise for applications in protein engineering and understanding viral evolution.