Generative AI-Assisted Molecular Design of AChEIs

Abstract Alzheimer’s disease (AD) remains a major neurodegenerative disorder with limited therapeutic options, while currently approved acetylcholinesterase inhibitors (AChEIs), such as donepezil, are associated with adverse effects including cardiotoxicity. Here, we integrated a deep learning-based framework to design novel AChEIs candidates with improved predicted cardiac safety. A reinforcement learning-guided GraphVAE (RL-GraphVAE) was employed for target-biased molecular generation. The generated compounds were prioritized through CardiotoxPred-based cardiotoxicity screening, molecular docking, triplicate molecular dynamics simulations, and chemical synthesis. Experimental evaluation identified D0209 as a lead molecule exhibiting a noncompetitive inhibition mechanism. Notably, compared with donepezil, D0209 showed ∼43-fold lower hERG channel inhibition, indicating an improved cardiac safety profile. Overall, this integrated computational and experimental workflow demonstrates the utility of generative modeling for the discovery of novel leads with improved predicted safety profiles, providing promising starting points for further optimization and biological evaluation.