7M-Multiomic Multiscale and Multifidelity Modeling via Machine Learning: Application to Diagnosis and to Macrovascular and Microvascular complications in NAFLD.

• Applicant: Professor Dr. Nikolaos Perakakis
• Subject Area: Endocrinology, Diabetology, Metabolism
• Term: from 2017 to 2019
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 389891681

Nonalcoholic fatty liver disease (NAFLD) has a prevalence of 30% in the general population, 40-70% in the obese and 60-90% in the diabetic population in industrialized countries. NAFLD is characterized by the presence of non-alcoholic fatty liver (NAFL) that can progress to steatohepatitis (NASH), fibrosis and cirrhosis. The survival of patients with NAFLD is reduced due to higher rates of hepatic and cardiovascular-related death. For this reason, the early diagnosis of NAFLD and the evaluation and treatment of the cardiovascular risk factors are very important. To date, liver biopsy is considered the gold-standard for the diagnosis of the disease, since non-invasive biomarkers and scores demonstrate low sensitivity or specificity. Additionally, the mechanisms associated with the increased cardiovascular mortality in NAFLD have not been adequately identified and described. Due to big differences in methodological approaches, it is still not clear whether and how NAFLD can increase the risk of vascular occlusion. Aims of the proposal are to develop a simplified algorithm with the minimum necessary parameters to diagnose NAFL, NASH with a sensitivity and specificity >90% (Aim 1), to investigate and compare biomechanics and rheology of blood (Aim 2) as well as the pathogenesis of macro- and microvascular disease (Aim 3) in NAFL and NASH. For Aim 1, the first-ever multiomic analysis including serum proteomics, metabolomics, lipidomics and glycomics in n=160 subjects with liver-biopsy (lean and overweight/obese subjects without NAFLD, overweight/obese patients with NAFL or NASH) will be performed. Data will be analyzed with advanced computational mathematical models to define an equation for the prediction of NAFL or NASH. For Aim 2 and Aim 3, a multiscale, multifidelity biomechanistic model for NAFLD will be developed. At the micro-level ex vivo, erythrocyte (RBC) deformability and rouleau formation, platelet size and rouleau formation and platelet-RBC interaction/aggregation will be evaluated. At the meso-macro level ex vivo changes in blood viscosity, plasma viscosity and thrombus formation will be investigated. At the macro-level, blood flow, thrombus formation and stability will be investigated in 3D-printed channels resembling patient-specific large vessels. For the formation of 3D-printed patient-specific channels in vivo data from carotid artery sonography and 24h-cardiovascular assessment system to imitate patient-specific atherosclerosis and arterial stiffness will be used. In all steps of the multiscale analysis, multifidelity modeling via deep Gaussian processes will be employed to discover "unknown" functional relationships and critical pathways and to bridge scales, synergistically from in vivo data, ex vivo data and simulations. To summarize, the current study aims to significantly improve the available methods for the non-invasive diagnosis of NAFLD as well as for the evaluation of vascular disease risk related to the disease.

• DFG Programme: Research Fellowships
• International Connection: USA
• Host: Professor Christos Mantzoros, Ph.D.