Redefining the CADD Paradigm: Harnessing Artificial Intelligence from Target Discovery to Lead Optimization

Abstract

Artificial intelligence (AI) is reshaping computer-aided drug design (CADD) by enabling data-driven analysis and generation across the early drug-discovery continuum. Conventional CADD approaches—including molecular docking, pharmacophore modelling, quantitative structure-activity relationship analysis and molecular dynamics—remain valuable, but their predictive performance can be constrained by simplified scoring functions, incomplete structural information and the enormous size of chemical space. Machine learning (ML), deep learning, graph neural networks, protein language models, geometric learning and generative models increasingly complement these methods. AI can support disease and target prioritization, protein-structure prediction, binding-site analysis, drug-target interaction prediction, virtual screening, de novo molecular generation, ADMET prediction, retrosynthetic planning and multi-parameter lead optimization. Recent advances such as AlphaFold3, deep-learning docking, ultra-large-library screening and generative molecular design have expanded the scope of structure-based and ligand-based discovery. Importantly, AI-generated candidates must be assessed for potency, selectivity, physicochemical properties, pharmacokinetics, toxicity, synthetic accessibility and uncertainty before experimental testing. Data leakage, bias, activity cliffs, distribution shift, limited prospective validation and interpretability remain significant barriers. This review presents an integrated AI-CADD framework from target identification to lead optimization and emphasizes the design-make-test-analyze cycle as the practical bridge between computational prediction and experimental medicinal chemistry.