IGT

IGT predicts drug–target binding activity and binding poses by modeling intermolecular topological and spatial information to improve virtual screening and drug discovery.

Key Features:

• Intermolecular graph representation: Represents molecules and proteins using intermolecular graph information that captures topological and spatial relationships.
• Three-way Transformer architecture: Employs a three-way Transformer-based architecture to integrate multiple interaction pathways between ligand and receptor representations.
• Dedicated attention mechanism: Uses a dedicated attention mechanism within the architecture to explicitly model intermolecular interactions.
• Joint activity and pose prediction: Produces predictions for both binding activity and binding pose for drug–target pairs.
• Improved predictive performance: Outperforms the second-best available models by 9.1% for binding activity prediction and by 20.5% for binding pose prediction.
• Generalization to unseen receptors: Demonstrates superior generalization when applied to unseen receptor proteins.
• Experimental validation (SARS-CoV-2): Identified active compounds against SARS-CoV-2 with 83.1% of predicted actives validated by wet-lab experiments and predicted poses closely matching native configurations.

Scientific Applications:

• Virtual screening: Prioritizes candidate compounds for virtual screening workflows in drug discovery.
• Binding activity prediction: Predicts ligand–receptor binding activity to support lead selection and prioritization.
• Binding pose prediction: Predicts binding poses to inform structure-based design and hypothesis generation.
• Discovery for emerging pathogens: Enables identification and prioritization of active compounds against targets such as SARS-CoV-2 for experimental follow-up.

Methodology:

Models intermolecular graphs using a three-way Transformer-based architecture with a dedicated attention mechanism to capture topological and spatial intermolecular interactions for joint binding activity and pose prediction.