Qianhui Liu, Xiaorong Tan, Feifan Xie, Defang Ouyang, Wenbin Zeng, Jie Dong
Peptide-based therapeutics hold substantial promise for treating diverse diseases, yet poor stability, limited permeability, rapid clearance and context-dependent behavior continue to hinder delivery and clinical translation. Today, artificial intelligence (AI) is increasingly used to support early property evaluation and enable rapid screening and prioritization of candidate peptides. However, given these peptide-specific pharmacokinetic challenges, accurate prediction of peptide absorption, distribution, metabolism, excretion and toxicity (ADMET) remains a key bottleneck, posing considerably greater challenges than conventional small-molecule ADMET prediction and therefore warranting a focused synthesis of recent advances. This review examines AI-driven peptide ADMET prediction, synthesizing methodological advances and limitations, including peptide-specific data scarcity, limited transferability and inadequate representations of modified peptides. It further outlines emerging directions, including interrelated ADMET modeling, class-specific peptide predictors and integration of AI with physiologically based pharmacokinetic models.