Generative AI for drug discovery and protein design: the next frontier in AI-driven molecular science
Generative artificial intelligence (AI) has emerged as a disruptive paradigm in molecular science, enabling algorithmic navigation and construction of chemical and proteomic spaces through data-driven modeling. This review systematically delineates the theoretical underpinnings, algorithmic architec...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | Medicine in Drug Discovery |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2590098625000107 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Generative artificial intelligence (AI) has emerged as a disruptive paradigm in molecular science, enabling algorithmic navigation and construction of chemical and proteomic spaces through data-driven modeling. This review systematically delineates the theoretical underpinnings, algorithmic architectures, and translational applications of deep generative models—including variational autoencoders (VAEs), generative adversarial networks (GANs), autoregressive transformers, and score-based denoising diffusion probabilistic models (DDPMs)—in the rational design of bioactive small molecules and functional proteins. We examine the role of latent space learning, probabilistic manifold exploration, and reinforcement learning in inverse molecular design, focusing on optimization of pharmacologically relevant objectives such as ADMET profiles, synthetic accessibility, and target affinity. Furthermore, we survey advancements in graph-based molecular generative frameworks, LLM-guided protein sequence modeling, and diffusion-based structural prediction pipelines (e.g., RFdiffusion, FrameDiff), which have demonstrated state-of-the-art performance in de novo protein engineering and conformational sampling. Generative AI is also catalyzing a paradigm shift in structure-based drug discovery via AI-augmented molecular docking (e.g., DiffDock), end-to-end binding affinity prediction, and quantum chemistry-informed neural potentials. We explore the convergence of generative models with Bayesian retrosynthesis planners, self-supervised pretraining on ultra-large chemical corpora, and multimodal integration of omics-derived features for precision therapeutics. Finally, we discuss translational milestones wherein AI-designed ligands and proteins have progressed to preclinical and clinical validation, and speculate on the synthesis of generative AI, closed-loop automation, and quantum computing in future autonomous molecular design ecosystems. |
|---|---|
| ISSN: | 2590-0986 |