GANs in Peptide Drug Discovery: From De Novo Design to Multi-Property Optimization and Clinical Translation

Peptide drugs are vital for treating intractable diseases, yet traditional discovery is limited by huge sequence space and poor pharmacokinetics. Generative Adversarial Networks (GANs) and related variants (CGAN, WGAN-GP, MPOGAN) are increasingly used as auxiliary computational tools for peptide drug discovery, primarily for de novo sequence exploration and multi-property optimization of antimicrobial, antiviral and anticancer peptides. Reported gains in predicted activity or novelty are study-specific and frequently remain limited to in silico evaluation or early in vitro assays. This review summarizes GAN architectures, database foundations, and diverse applications; discusses major limitations, including multi-label data scarcity and property trade-offs; and proposes future directions, including LLM-assisted annotation and closed-loop AI-experimental platforms. It aims to organize current evidence for AI-driven peptide design, emphasize methodological and translational limitations, and outline a realistic pathway toward clinical translation.