Abstract

The conventional pharmaceutical research and development paradigm is characterized by protracted timelines exceeding a decade, financial burdens surpassing $2.6 billion per approved therapeutic, and attrition rates exceeding 90% during clinical development, primarily attributable to suboptimal pharmacokinetics, unforeseen toxicity, and inadequate efficacy compounded by heterogeneous patient populations. Artificial intelligence has emerged as a transformative technological paradigm capable of addressing these interconnected challenges through data-driven decision-making, computational extrapolation of structure-activity relationships, and probabilistic modeling of complex biological systems. This review systematically examines the integration of artificial intelligence across the entire pharmaceutical development continuum, encompassing machine learning, deep learning, generative adversarial networks, reinforcement learning, natural language processing, and emerging adaptive learning architectures including liquid neural networks and digital twins. Key applications analyzed include target discovery through multi-omic data integration, de novo molecular design with generative algorithms, high-throughput virtual screening, predictive absorption-distribution-metabolism-excretion-toxicity modeling employing graph neural networks and automated machine learning, drug repurposing via knowledge graph inference, and clinical trial optimization through explainable artificial intelligence-driven patient enrichment and synthetic control arm construction. Representative case studies including the artificial intelligence-designed rentosertib (ISM001-055) demonstrating positive Phase IIa outcomes in idiopathic pulmonary fibrosis and the deep learning-identified antibiotic halicin are critically evaluated alongside discontinued candidates to provide balanced translational perspectives. Persistent challenges encompassing data heterogeneity, algorithmic interpretability, regulatory qualification, and ethical deployment are systematically addressed, concluding that artificial intelligence constitutes an indispensable complement to traditional methodologies rather than a wholesale replacement, with autonomous agentic systems and federated learning architectures poised to define the next developmental epoch.