Geometric Graph Learning to Predict Changes in Binding Free Energy and Protein Thermodynamic Stability upon Mutation

Area of Science:

• Computational biology
• Biophysics
• Machine learning

Background:

• Predicting changes in binding free energy due to mutations is crucial for protein engineering and drug development.
• Current machine learning methods face challenges in accuracy and generalization for these predictions.

Purpose of the Study:

• Introduce Geometric Graph Learning for Protein-Protein Interactions (GGL-PPI), a novel method for predicting mutation-induced binding free energy changes.
• Evaluate GGL-PPI's performance and generalization capabilities on standard datasets and protein stability prediction.

Main Methods:

• Utilized atom-level graph coloring and multiscale weighted colored geometric subgraphs to represent biomolecular structures.
• Integrated geometric graph representation with machine learning algorithms.

Main Results:

• GGL-PPI demonstrated superior performance on AB-Bind, SKEMPI 1.0, and SKEMPI 2.0 datasets.
• The model showed strong generalization by accurately predicting protein thermodynamic stability on a blind test set for both direct and reverse mutations.

Conclusions:

• GGL-PPI offers a precise and generalizable approach for predicting mutation-induced changes in binding free energy and protein stability.
• This advancement provides valuable insights for protein complex comprehension and accelerates drug design efforts.