Artificial Intelligence in Medicinal Chemistry Decision-Making: From Virtual Screening to Lead Optimization

Artificial Intelligence (AI) is increasingly embedded within modern medicinal chemistry workflows, particularly across the Design-Make-Test-Analyse (DMTA) cycle, where critical decisions must be made under conditions of uncertainty. Rather than functioning as a stand-alone predictor, AI is best understood as a decision-support framework that assists chemists in selecting screening hits, prioritising analogues for synthesis, managing multi-parameter optimisation, and identifying developability and safety risks at early stages. This structured narrative review examines how contemporary Machine Learning (ML) and Deep Learning (DL) approaches are being applied at key decision points in small-molecule drug discovery, including virtual screening and hit triage, Structure-Activity Relationship (SAR) prioritisation, lead optimisation for Absorption, Distribution, Metabolism, Excretion and Toxicity (ADMET) liabilities, and synthetic feasibility assessment through retrosynthesis planning. By integrating ligand-based and structure-based models, multi-task learning, active learning, generative design, and uncertainty-aware predictions, AI systems can reduce experimental burden while improving the quality of decisions made by medicinal chemists. However, the practical impact of these tools depends critically on appropriate validation strategies, applicability domain assessment, and probability calibration to avoid misleading confidence. Taken together, the evidence reviewed here supports models of AI-augmented medicinal chemistry, in which data-driven predictions and human chemical intuition jointly enable more efficient, transparent, and reliable progression from hits to optimised lead candidates.