Possibilities and limitations of artificial intelligence in food-derived peptides

The deep integration of artificial intelligence (AI) with research on food-derived peptides (FDPs) signifies a paradigm shift towards intelligent precision design and functional optimization in developing FDP products. Random forests, convolutional neural networks and other machine learning and deep learning technologies have markedly expedited bioactive peptide discovery, functional characterization and structure-activity relationship elucidation, while their comprehensive application confronts critical challenges encompassing dataset integrity, model architecture optimization, interpretability constraints and experimental verification requirements. Consequently, a comprehensive understanding of the existing limitations and future opportunities of AI technology in the field of FDPs is required. This review evaluates traditional methodologies and AI applications in FDP research, addressing current limitations and prospects. AI faces several challenges in predicting complex protein structures and FDPs. Dimensionality curse and local optima during protein structure prediction reduce accuracy. In FDP prediction, the selection of amino acid descriptors limits model stability. Additionally, the scarcity of FDP datasets leads to accuracy reduction in AI model training and introduces class distribution bias. Traditional cross-entropy loss further exacerbates class distribution skew. These issues collectively constrain the application of AI in FDP prediction. AI enhances peptide screening efficiency through large-scale data processing, reducing research costs. Standardized multidimensional databases integrated with heterogeneous AI architectures improve predictive model accuracy and cross-domain adaptability. Computational visualization frameworks elucidate bioactive determinants and quantitative structure-activity relationships. Metabolomic-gut microbiota interdisciplinary strategies enable the precision design of bioactive peptides for personalized nutrition. AI-optimized enzymatic hydrolysis processes and high-throughput validation platforms advance mechanistic understanding and industrial production, despite the persisting challenges. © 2025 Society of Chemical Industry.