Machine learning and mathematical modeling for precision medicine in metabolic diseases

Type 2 diabetes (T2D), Non-Alcoholic Fatty Liver disease (NAFLD) and steatohepatitis (NASH) are metabolic diseases whose global prevalence is reaching epidemic proportions and parallels the incidence of obesity. Often coexisting, these heterogeneous diseases are associated with increased insulin resistance and alterations of hepatic and extra-hepatic glucose and lipid metabolism, as well as a higher prevalence of other factors/comorbidities like hypertension, dyslipidemia or abdominal fat accumulation. The causal mechanisms linking all these factors are poorly understood, but adipose tissue dysfunction and lipolysis seem to play a major role in the pathophysiology of both NAFLD and T2D. On the other hand, alterations in glucose production and disposal that are the most important determinants of hyperglycemia in T2D are poorly investigated in NAFLD. Nowadays no pharmacological treatment has been approved for NAFLD/NASH and, although some antihyperglycemic drugs targeting adipose tissue, such as peroxisome proliferator-activated receptor γ (PPAR-γ) agonists, have shown some efficacy, lifestyle modification remains key in the clinical management of NASH and in the prevention and the delay of T2D. Machine learning and mathematical modelling have emerged as fundamental tools for the study of metabolic diseases. By managing dataset of large size and at the same time integrating different layers of biological knowledge, like clinical and ‘omics’ data, these tools allow to investigate the diseases with a systemic approach and their application has been proved useful to fulfill classification and prediction tasks, but also to unravel hidden biochemical patterns. The aim of this thesis was to investigate the alterations of glucose metabolism and its association with adipose tissue insulin resistance in subjects with metabolic diseases like NAFLD and T2D (or at high risk to develop them). To this purpose, I applied mathematical modeling and machine learning to investigate pathophysiological mechanisms associated with glucose and lipid metabolic dysregulations (Chapter 2) and to evaluate the heterogeneity and the inter-individual variability in response to intervention targeting adipose tissue metabolism (Chapter 3). I found that hepatic glucose production and in particular the metabolic flux through gluconeogenesis were increased in subjects with NASH and severe fibrosis and resulted mainly from excessive substrates availability. Indeed, an important role is played by increased visceral fat depot and adipose tissue dysfunction that were associated with increased hepatic insulin resistance and severe hepatic necroinflammation, even in individuals without obesity or T2D. Improvement in adipose tissue metabolism and its re-distribution was also related to the beneficial effect of insulin-sensitizing pharmacological (i.e., pioglitazone) and non-pharmacological interventions in subjects with or at high risk of metabolic diseases. Treatment with the PPAR- γ agonist pioglitazone improved liver histology and glucose metabolism in subjects with NASH mainly through the reduction of visceral fat and the improvement of adipose tissue insulin resistance. Lifestyle intervention, through dietary counselling and increased physical activity, contributed to improve glucose metabolism primarily in those that had high insulin resistance with high glucose-stimulated insulin response at baseline. These findings further elucidate the mechanisms that regulate glucose and lipid metabolism potentially responsible for the development of T2D and NAFLD and identify the metabolic changes obtained with pharmacological and non-pharmacological interventions that could be used towards personalized interventions in a precision medicine approach.