MORLD

MORLD applies reinforcement learning to directly modify ligand structures based on a target protein structure to optimize predicted binding affinity for lead discovery without requiring additional training data.

Key Features:

• Structure-guided ligand modification: directly alters ligand structures using a given target protein structure.
• Training-free optimization: operates without requiring additional training data.
• Reinforcement learning optimization: uses reinforcement learning algorithms that iteratively adjust molecular configurations.
• Docking-based affinity evaluation: maximizes predicted binding affinity via docking simulations.
• Autonomous chemical space exploration: explores vast chemical space to identify promising lead compounds.
• Demonstrated rapid generation: generated DDR1 kinase inhibitors in under two days on moderate compute and produced D4 dopamine receptor (D4DR) agonists from scratch from ultra-large compound libraries.

Scientific Applications:

• Lead optimization: improve predicted binding affinity and molecular configurations for lead compounds.
• De novo ligand generation: generate novel agonists and inhibitors from scratch.
• Kinase inhibitor discovery: applied to discoidin domain receptor 1 kinase (DDR1) inhibitor generation.
• GPCR ligand discovery: applied to D4 dopamine receptor (D4DR) agonist generation.
• Alternative to virtual screening: bypass traditional virtual screening for ultra-large compound libraries using direct modification and docking-guided reinforcement learning.

Methodology:

MORLD uses reinforcement learning algorithms to iteratively adjust ligand configurations based on a given target protein structure, maximizing predicted binding affinity via docking simulations while operating without additional training data.