Diffusion Models for Small-Molecule Design

Abstract Diffusion models (DMs) have emerged as a class of generative models for controllable small-molecule design in drug discovery capable of progressively transforming noisy or partially specified molecular states into candidate structures under structural, geometric, and property constraints. This Review provides a comprehensive overview of the methodological foundations and representative applications of diffusion-based molecular generation, emphasizing the alignment among molecular representations, generative dynamics, controllable design, and evaluation and validation. Particular attention is given to recent advances in property- and structure-guided generation, target-aware and fragment-based design, multiobjective optimization, and synthesis-aware generation. We also discuss the roles of data sets, computational evaluation, and prospective experiments in assessing the chemical and biological relevance of generated molecules. By organizing these developments within a unified conceptual framework, we provide computational chemists and drug discovery researchers with a structured roadmap that balances the mathematical rigor with chemical intuition. Throughout, we highlight current methodological limitations and unresolved challenges and conclude with future directions toward chemically credible and experimentally grounded molecular design. To facilitate navigation of this rapidly evolving field, we introduce an interactive companion platform (https://diffsmd.scbdd.com) that provides a curated, continuously updated repository of model architectures and visualizations.