ET-Score

ET-Score predicts ligand-protein binding affinity using an Extremely Randomized Trees (Extra Trees) machine-learning scoring function for molecular docking simulations.

Key Features:

• Machine Learning Integration: ET-Score uses the Extremely Randomized Trees (Extra Trees) algorithm to model structure–affinity relationships.
• Feature Engineering: Uses distance-weighted interatomic contacts between atom type pairs from the ligand and the protein as input features.
• Performance Metrics: Evaluated on the PDBbind 2016v core set with Pearson's correlation coefficient 0.827 and RMSE 1.332.
• Computational Efficiency: Maintains an extremely low computational cost for prediction.
• Comparative Advantage: Benchmarked against other ML-based scoring functions and classical scoring methods, outperforming most ML models and all classical approaches.

Scientific Applications:

• Computational drug design: Provides quantitative affinity predictions to support prioritization of potential therapeutic candidates.
• Molecular docking scoring: Serves as a machine-learning scoring function to improve ranking of docked ligand poses.

Methodology:

Implements the Extremely Randomized Trees algorithm using distance-weighted interatomic contact features between ligand and protein atom type pairs and was evaluated on the PDBbind 2016v core set (reported Pearson's r = 0.827, RMSE = 1.332).