Artificial intelligence in food and nutrition science: a paradigm-centric review of computational frameworks and...

Precision nutrition on a global scale necessitates an understanding of food not as static collections of so-called macronutrients but rather as dynamic and heterogeneous biochemical matrices. These complex non-linear interactions between food composition, gastrointestinal digestion and the human microbiome are difficult to capture using traditional empirical experimental methods. This review articulates a paradigm-based framework that reconceptualizes artificial intelligence (AI) in food and function science, from generic industrial applications to the computational modeling of physiological and biochemical phenomena. We specifically explore critical integrations of Physics-Informed Neural Networks (PINNs) with established data-driven methods to circumnavigate the epistemological limitations of entirely data-driven models within standardized frameworks (e.g., INFOGEST), assessing their use in simulating gastrointestinal mass transfer and dissolution kinetics to achieve predictive accuracies up to R² = 0.91 in complex protein digestibility matrices. Finally, we demonstrate how Microbial Community-scale Metabolic Modeling (MCMM) and multimodal machine learning mechanistically couple specific dietary inputs with unique microbiome responses, yielding up to 95% diagnostic accuracy in differentiating diet-responsive metabolic states and individualized postprandial glycemic outcomes. We also emphasize the importance of Graph Neural Networks (GNNs) in rationally designing bioactive peptides, and Artificial Neural Networks (ANNs) for the optimization of microencapsulation, which have demonstrated the capacity to reduce physical experimental formulation trials by over 60%. Furthermore, we evaluate AI's emerging role in precision aquaculture, where computer vision and predictive models yield 94-99% accuracy in disease detection and reduce feed conversion ratios by up to 11%. This review discusses the need to connect computational intelligence and food function, delineating a path toward autonomous metabolically-aware food ecosystems whilst identifying important research gaps in mechanistic interpretability, data standardization, and cultural bias underpinning precision nutrition.