SBMolGen

SBMolGen generates de novo small molecules by integrating a recurrent neural network (RNN), Monte Carlo Tree Search (MCTS), and docking simulations to propose 3D-structure-aware candidates with improved binding affinity to protein targets.

Key Features:

• Deep Learning Integration: Employs a recurrent neural network (RNN) to generate novel molecular structures from sequential representations.
• Monte Carlo Tree Search (MCTS): Uses MCTS to explore chemical space systematically and evaluate candidate molecular configurations for novelty and binding potential.
• Docking Simulations: Performs docking simulations to predict and refine 3D poses of generated molecules bound to specific protein targets and assess interactions.
• Enhanced Binding Affinity Prediction: Produces molecules that evaluation results indicate can achieve superior binding affinity scores compared to known active compounds.
• Broad Chemical Space Exploration: Generates a diverse array of molecular structures to expand chemical space beyond traditional methods.

Scientific Applications:

• Structure-based de novo drug design: Generates 3D-structure-aware candidate molecules for use in rational drug design workflows.
• Targeted inhibitor discovery: Applies to discovery of binders against complex protein families such as kinases and G protein-coupled receptors (GPCRs).
• Lead generation and optimization: Produces novel binding-active molecules suitable for downstream lead optimization and preclinical development stages.

Methodology:

The RNN generates candidate molecules, MCTS evaluates and expands promising candidates, and docking simulations assess and refine 3D poses and binding affinities.