Machine Learning-Driven Approaches for Identification of Natural Bioactive Peptides: Recent Advances, Challenges, and Future Perspectives

Abstract

Natural bioactive peptides (NBAPs) have gained increasing prominence in functional food development due to their health-promoting properties. However, conventional experimental identification of NBAPs is often time-consuming and laborious, highlighting the need for efficient computational approaches. Machine learning (ML) has emerged as a powerful computational approach for the discovery of NBAPs by enabling peptide sequence analysis, feature representation, activity prediction, and candidate prioritization. This review summarizes recent advances in ML-driven identification of NBAPs, with an emphasis on feature encoding strategies, representative algorithms, model evaluation methods, and publicly available prediction tools. The applications of ML approaches in representative bioactive peptide categories, including anticancer, antihypertensive, antimicrobial, antioxidant, and metabolic regulatory peptides, are systematically discussed. Furthermore, current challenges associated with limited and biased datasets, inconsistent negative sample construction, insufficient model interpretability, and poor cross-dataset generalization are highlighted. Finally, future perspectives are discussed, particularly the development of structure-aware deep learning strategies that integrate peptide 3D information with advanced representation learning to improve the understanding of structure–activity relationships. This review provides comprehensive insights into the current status and future directions of ML-driven bioactive peptide identification.