ProS-GNN

ProS-GNN predicts changes in protein stability caused by mutations using deep graph neural networks to model structure–property relationships for applications in drug development and immunotherapy.

Key Features:

• Graph Neural Network Architecture: Encodes molecular structure–property relationships using a GNN with message passing to capture dependencies within protein structures.
• Spatial Insights: Incorporates raw atom coordinates to represent three-dimensional molecular geometry for stability prediction.
• Bias and Overfitting Mitigation: Addresses overfitting on training data and anti-symmetric biases between direct and reverse mutations to improve robustness.
• Efficiency: Performs predictions with ultra-low time consumption suitable for large-scale analyses.

Scientific Applications:

• Drug Development: Predicts mutation-induced stability changes to aid understanding of drug failure mechanisms.
• Immunotherapy Strategies: Provides insights into protein stability alterations caused by mutations relevant to immunotherapy design.
• Case Studies: Applied to predict the Gibbs free energy change for Pyrazinamide as an example application.

Methodology:

Performs mutant-part data extraction, encodes molecular structures using message passing within a GNN framework incorporating raw atom coordinates, and was trained on the S2648 and S3412 datasets with validation against the Ssym and Myoglobin datasets.