Artificial intelligence in the design of functional food products: opportunities and limitations of digital screening of bioactive compounds

Abstract

The review summarizes current approaches to the application of artificial intelligence for the discovery, selection and preliminary technological assessment of bioactive compounds in the design of functional food products. Machine learning, QSAR/QSPR modeling, virtual screening, docking combined with machine learning, graph-based models and explainable artificial intelligence are considered as tools for accelerating early-stage selection of bioactive peptides, polyphenols, food and microbial metabolites, probiotics and prebiotic systems. The paper emphasizes that computational prediction is not evidence of ingredient or product efficacy and must be followed by biochemical, technological, toxicological and regulatory validation. Particular attention is paid to bakery, cereal, flour confectionery and fermented food products, where a bioactive component must be food-relevant, technologically stable, safe and organoleptically acceptable. It is emphasised that the quality and comparability of source data, together with the complexity of the food matrix, remain key limitations, and that models transferred directly from pharmacological chemistry into food technology require caution. For bakery, cereal, flour confectionery and fermented products, particular attention is paid to the stability of bioactive components during mixing, fermentation, baking, drying and storage, their interaction with matrix components, as well as sensory properties and acceptable dosage. Open chemical, food, metabolomic and protein resources, molecular descriptors and model architectures are discussed as parts of a single technological chain. Artificial intelligence is therefore positioned as a tool for hypothesis generation and prioritisation rather than as an independent source of proof of efficacy.