Integrated machine learning and mechanistic modelling of Metabolic Syndrome development and dynamics

Metabolic derailments associated with metabolic syndrome and type 2 diabetes can be studied with mixed meal tests (MMT’s). The plasma metabolome enriched with measurements of other biomarkers, such as hormones and cytokines, provide valuable information about the physiological state of an individual. An increasingly important, but complex task is to extract biomedical parameters with diagnostic value from large and multivariate datasets. We applied model-based data processing and analysis, combining computer simulation models of the human physiological system, stochastic models of uncertainties and machine learning of time-series data obtained from repeated blood sampling during MMT’s.
In a mouse model of metabolic syndrome we identified differences in lipid metabolism to be associated with variation in weight gain and development of NAFLD (fatty liver disease). The computational model predicted the progression of dyslipidemia to be linked to bile acids, which was confirmed in a validation study including a larger group of mice that were followed for a longer period of time.
To investigate the role of bile acids in humans with metabolic syndrome a detailed simulation model of bile acid metabolism and physiology was developed. The dozens of different bile acid species present in blood are to a large extent produced by gut bacteria. Model-based analysis of plasma bile acids provides a metabolic ‘window’ on the gut microbiome and other digestive processes in the gastrointestinal tract. The model was applied to simulate bariatric surgery in patients with metabolic syndrome. The model predicts changes in bile acid concentrations and dynamics in the small intestine to result in a stronger and faster GLP-1 response, hence insulin secretion, explaining observations of rapid glycemic improvement after surgery. The simulation model turned out to be sufficiently robust that personalized variants for individual patients could be made, using MMT plasma bile acid metabolomics as input.