Strategy and efficiency-redefined discovery of novel nanomolar tyrosinase inhibitors: AI de novo molecular generation + expert-guided structural optimization

Artificial intelligence (AI) has played an excellent supporting role in novel drug discovery and development. This study introduces a reinforcement learning (RL) model based on the Soft Actor-Critic (SAC) algorithm for AI-driven de novo molecular generation targeting tyrosinase. The model facilitates forward molecular generation design by integrating a chemical reaction template and a molecular building block library, concurrently performing molecular docking and assessing drug-likeness. Through sequential decision-making, signal feedback, and a dynamic learning process, the model generates molecules exhibiting potent target affinity, optimal drug-like properties, and good synthetic feasibility. The AI-generated molecules undergo rigorous manual screening, synthesis, and biological evaluation, culminating in the identification of a prioritized lead compound V. Subsequent structural optimization of compound V reveals a series of compounds with significantly enhanced activity, shifting inhibitory potency from the micromolar to the nanomolar range. The optimized compound, V-24, demonstrates low cytotoxicity and significant anti-melanogenic activity both in cell melanogenesis inhibition and zebrafish anti-pigmentation models. Notably, it effectively reduces melanin content in an ultraviolet light-induced human 3D skin pigmentation model, exhibiting the potential to serve as a promising tyrosinase inhibitor for the treatment of skin pigmentation. More importantly, this "AI de novo Molecular Generation + Expert-Guided Structural Optimization" work demonstrates that integrating an AI algorithm with traditional medicinal chemistry experience is a novel approach and efficiency-redefined strategy for drug discovery.