Artificial intelligence-driven drug discovery: a deep learning paradigm shift in pharmaceutical research and development

Abstract

Efficient drug discovery is essential to mitigate the high attrition rates and capital-intensive nature of traditional pharmaceutical research. Deep learning (DL) has catalyzed a paradigm shift, offering unprecedented computational capabilities to accelerate this process. This review systematically summarizes advancements in DL methodologies for drug discovery. First, the critical data foundations, Molecular Representation Learning, and core architectures underpinning artificial intelligence (AI)-driven therapeutics are elucidated, with a particular emphasis on the evolution of generative AI frameworks (such as Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and diffusion models) for chemical space exploration. Subsequently, diverse DL applications across the discovery pipeline are investigated, encompassing target identification, virtual screening, de novo molecular design, retrosynthetic analysis, absorption, distribution, metabolism, excretion, and toxicity (ADMET) profiling, and drug repurposing. Furthermore, persistent challenges, particularly data scarcity, model opacity, experimental validation bottlenecks, and escalating computational resource requirements, are critically evaluated. To bridge the gap between in silico predictions, chemical synthesis, and biological reality, emerging solutions such as federated learning, causality-aware Explainable AI (XAI), closed-loop autonomous laboratories, and efficiency-oriented algorithmic strategies are explored. Ultimately, this article provides strategic insights into leveraging DL, underscoring its transformative potential in driving next-generation medicinal and therapeutic innovation.