Mol-AIR: Molecular Reinforcement Learning with Adaptive Intrinsic Rewards for Goal-Directed Molecular Generation.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence

Background:

• Optimizing molecular structures with desired properties is key in AI-driven drug discovery.
• Deep generative models combined with reinforcement learning are effective but struggle with chemical space exploration and property optimization.
• Existing methods face limitations in efficiently generating molecules with specific target properties.

Purpose of the Study:

• To introduce Mol-AIR, a novel reinforcement learning framework designed for goal-directed molecular generation.
• To enhance the exploration of vast chemical spaces and optimize specific molecular properties.
• To provide a more efficient pathway for discovering novel therapeutic compounds.

Main Methods:

• Mol-AIR utilizes a reinforcement learning-based framework with adaptive intrinsic rewards.
• The framework integrates history-based and learning-based intrinsic rewards.
• It employs random distillation networks and counting-based strategies for reward optimization.

Main Results:

• Mol-AIR demonstrated superior performance compared to existing methods in benchmark tests.
• The framework successfully generated molecules with desired properties, including penalized LogP, QED, and celecoxib similarity.
• Performance was achieved without requiring prior knowledge of the target molecules.

Conclusions:

• Mol-AIR represents a significant advancement in AI-based drug discovery.
• The framework offers a more efficient and effective approach to goal-directed molecular generation.
• Mol-AIR accelerates the discovery of novel therapeutics with optimized properties.