AI-Guided Molecular Design Using Deep Reinforcement Learning for Drug Discovery

Abstract

Drug discovery is a time-consuming and resource-intensive process, with many candidate compounds failing in early development stages. A key challenge is efficiently identifying chemically valid and pharmacologically relevant molecules within a vast chemical space, particularly for cancer-targeted and disease-specific applications. To address this, we present an artificial intelligence framework for de novo molecular design using deep reinforcement learning, specifically a Deep Q-Network (DQN) model, as introduced in our 2025 conference paper [1]. The framework leverages molecular descriptors derived from the ZINC dataset and formulates molecule generation as a sequential decision-making process. By expanding the action space, the model enhances exploration of chemical space while optimizing drug-likeness properties. The generated compounds demonstrate favorable physicochemical characteristics, including balanced LogP values, appropriate molecular weight, and acceptable hydrogen-bonding profiles. Notably, the framework achieves 100% chemical validity for generated molecules. Compared with existing approaches, the proposed method offers competitive performance while maintaining a simpler and more interpretable architecture. These findings align with broader advancements in artificial intelligence and reinforcement learning for biomedical and health applications, including oncology, disease detection, and treatment optimization [2]–[4].