Mathematical modeling in precision nutrition

Abstract: Precision nutrition (PN) aims to tailor diets for individuals or groups based on comprehensive data to improve prevention of diseases, such as cardiometabolic diseases. Predicting individual postprandial metabolic responses and identifying individuals with similar metabolic phenotypes (metabotypes) could guide tailored diet strategies. While many metabolic markers are associated with health outcomes, predictive methods for high-dimensional postprandial responses are lacking. Furthermore, metabotyping has mainly been performed using cluster analysis on data from static blood markers or from responses to single dietary challenges. However, methods to incorporate time-resolved data from several dietary challenges or multi-omics (e.g., metabolomics and microbiomics) have not been explored properly. This thesis breaks ground by addressing these challenges using time-resolved and static metabolomics, gut microbiota, dietary, and health status data. The research presented in this thesis showed successful identification of metabotypes related to different cardiometabolic risks in a free-living population using multiple factor analysis of static microbiota and metabolomics. This led to deeper metabolic characterization compared to using single omics. Furthermore, dynamic mode decomposition was used to investigate the predictability of postprandial metabolic responses using the baseline metabolome and nutritional information of meals. The method was shown to be predictive in both measured (R2=0.4) and simulated (R2=0.65) data. It was also used along with the tensor decomposition CANDECOMP/PARAFAC to identify metabotypes relating to amino acid absorption in data from a crossover intervention study using repeated measurements from multiple dietary challenges, showing the utility of performing the two important PN tasks in one method. Finally, kinetic model parameters derived from postprandial plasma glucose dynamics were investigated to identify differential responders to meal challenges. Identified clusters were differently associated with type-2 diabetes risk markers and gut microbiota, which showed that differences in postprandial dynamics relate to type 2 diabetes risk markers and can be used to identify individuals at risk. In conclusion, the analytical methods developed in this thesis present a versatile toolbox that may be used to improve metabotyping in complex study designs, enable dynamic predictions of postprandial responses, and demonstrate the utility of postprandial dynamics in detecting individuals at risk of disease.