Generative Artificial Intelligence in De Novo Drug Design: Current Advances, Challenges, and Future Perspectives
Keywords:
AI-driven molecular optimization, Diffusion models, Generative Adversarial Networks (GANs), Generative Artificial Intelligence, Large Language Models (LLMs), Molecular generation, Reinforcement learning, Transformer models, Variational Autoencoders (VAEs)Abstract
Drug development is still a difficult, expensive, and time-consuming process with high attrition rates, underscoring the need for creative computational methods. The objective of this review is to objectively assess the latest developments, uses, difficulties, and prospects of Generative Artificial Intelligence (GenAI) in de novo drug creation and pharmaceutical innovation. To find pertinent publications from 2018 to 2026, a thorough literature search was carried out using PubMed, Scopus, Web of Science, ScienceDirect, IEEE Xplore, and Google Scholar. Variational autoencoders (VAEs), Generative Adversarial Networks (GANs), diffusion models, transformer architectures, reinforcement learning, and large language models (LLMs) are among the generative AI techniques that have been the subject of critical analysis and synthesis. According to the examined data, GenAI has greatly accelerated molecule creation, lead optimisation, target identification, and pharmacokinetic and pharmacodynamic property prediction while extending chemical space exploration beyond traditional techniques. The translational potential of these technologies has been demonstrated by the advancement of several AI-designed drug candidates into preclinical and clinical development. Widespread clinical deployment is still constrained by issues with data quality, model interpretability, synthetic feasibility, experimental validation, regulatory acceptance, ethical constraints, and intellectual property. Overall, by facilitating quicker, more effective, and data-driven pharmaceutical research, generative artificial intelligence is revolutionising de novo drug creation. To enable the safe and efficient transfer of AI-generated drug candidates into clinical practice, future advancements will rely on the integration of explainable AI, high-quality datasets, interdisciplinary cooperation, standardised validation procedures, and regulatory guidance.