DeepBindRG

DeepBindRG predicts binding affinity of protein-ligand complexes using a deep neural network that accounts for solvent effects, entropy changes, and multibody interactions.

Key Features:

• Deep Neural Network-Based Model: Employs a deep neural network architecture that implicitly learns binding mode and specificity of protein-ligand interfaces.
• Comprehensive Data Processing: Preserves critical interface contact information during data processing to provide optimized inputs for the model.
• Enhanced Predictive Accuracy: Validated on three independent datasets with RMSE ≈ 1.6–1.8 and R ≈ 0.5–0.6, outperforming AutoDock Vina which showed RMSE ≈ 2.2–2.4 and R between 0.42–0.57.
• Performance on Challenging Datasets: Demonstrated effective predictions on four challenging datasets from the DUD.E database where no experimental protein-ligand complexes were available.
• Comparative Evaluation: Compared with AutoDock Vina and the 4D-based deep learning method pafnucy to assess advantages and limitations.

Scientific Applications:

• Drug Discovery and Virtual Screening: Provides binding affinity predictions to support identification and prioritization of potential therapeutic compounds.
• Protein-Ligand Interaction Analysis: Enables study of interface-specific effects including solvent and entropy contributions in binding affinity estimation.

Methodology:

Training a deep neural network on large-scale protein-ligand datasets using preserved interface contact information as input to implicitly learn binding mode and specificity.