FRAIL: fragment-based reinforcement learning for molecular design and benchmarking on fatty acid amide hydrolase 1...

We propose FRAIL (Fragment-based Reinforcement Learning for Inhibitors), a generative AI framework that integrates fragment-based molecular design, multi- objective reinforcement learning, and molecular modeling to accelerate inhibitor discovery. Several deep generative models were fine-tuned on FAAH-1 (Fatty Acid Amide Hydrolase 1)-specific dataset and systematically benchmarked, with the best-performing model incorporated into FRAIL. The framework employs a customized reward function that jointly optimizes physicochemical properties and predicted bioactivity (pIC₅₀) to guide molecular generation toward FAAH- favorable chemotypes. FRAIL generated structurally novel, fragment-grown compounds exhibiting high predicted binding affinity, desirable drug-likeness, and synthetic accessibility. These findings demonstrate FRAIL's capability to enhance rational drug design and provide a reproducible pipeline for the discovery of experimentally viable FAAH inhibitors. Our pipeline source code is released in https://github.com/AppliedAI-Lab/FRAIL .