Artificial Intelligence-enhanced Computer-aided Drug Design: Recent Advances, Applications and Challenges

Abstract

CADD has, for four decades, provided the computational scaffolding on which rational drug discovery is built, encompassing structure-based and ligand-based approaches such as molecular docking, molecular dynamics, pharmacophore modelling, and QSAR analysis. Over the last decade this discipline has been transformed by the infusion of AI, giving rise to what is increasingly termed AI-driven drug design (AIDD). Deep neural networks, generative models, graph neural networks, and transformer-based language models now support nearly every stage of the discovery pipeline, from target identification and highly accurate protein structure prediction, through de novo molecular generation and virtual screening of ultra-large chemical libraries, to the prediction of ADMET properties and the optimisation of clinical trial design. This review synthesises recent advances in AI-enhanced CADD, covering key methodological families (generative adversarial networks, variational autoencoders, diffusion models, graph neural networks, and transformers), landmark applications including AlphaFold and its successors, deep-learning-guided antibiotic and antifibrotic drug discovery, and AI-designed molecules that have advanced into clinical trials. It further examines the persistent challenges that temper enthusiasm for the field — data scarcity and bias, limited model interpretability, poor generalisation beyond training distributions, weak experimental validation, and unresolved regulatory and ethical questions — and outlines emerging directions, including multimodal foundation models, physics-informed learning, and closed-loop autonomous laboratories, that are likely to shape the next generation of AI-enhanced drug discovery.