The paper introduces ReLeaSE (Reinforcement Learning for Structural Evolution), a novel computational strategy for de-novo drug design. ReLeaSE integrates two deep neural networks—generative and predictive—trained separately but used jointly to generate novel chemical libraries with desired properties. The generative model, a stack-augmented recurrent neural network (Stack-RNN), generates chemically feasible molecules, while the predictive model forecasts the desired properties of these compounds. The method is trained in two stages: first, both models are trained separately using supervised learning, and then they are jointly trained with reinforcement learning to bias the generation of new chemical structures towards those with the desired properties. The proof-of-concept study demonstrates the effectiveness of ReLeaSE in designing chemical libraries biased towards structural complexity, specific ranges of physical properties (e.g., melting point, hydrophobicity), and novel inhibitors of JAK2. The approach is versatile and can be applied to generate targeted chemical libraries optimized for single or multiple properties. The results show that ReLeaSE can produce a large number of novel and chemically valid compounds, with high synthetic accessibility, and accurately predict their properties.The paper introduces ReLeaSE (Reinforcement Learning for Structural Evolution), a novel computational strategy for de-novo drug design. ReLeaSE integrates two deep neural networks—generative and predictive—trained separately but used jointly to generate novel chemical libraries with desired properties. The generative model, a stack-augmented recurrent neural network (Stack-RNN), generates chemically feasible molecules, while the predictive model forecasts the desired properties of these compounds. The method is trained in two stages: first, both models are trained separately using supervised learning, and then they are jointly trained with reinforcement learning to bias the generation of new chemical structures towards those with the desired properties. The proof-of-concept study demonstrates the effectiveness of ReLeaSE in designing chemical libraries biased towards structural complexity, specific ranges of physical properties (e.g., melting point, hydrophobicity), and novel inhibitors of JAK2. The approach is versatile and can be applied to generate targeted chemical libraries optimized for single or multiple properties. The results show that ReLeaSE can produce a large number of novel and chemically valid compounds, with high synthetic accessibility, and accurately predict their properties.