M3-NAFLD: Microbiome-based Machine Learning Models for better Diagnosis of Non-Alcoholic Fatty Liver Disease

• Applicant: Professor Gianni Panagiotou, Ph.D.
• Subject Area: Gastroenterology ; Bioinformatics and Theoretical Biology ; Microbial Ecology and Applied Microbiology
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 544831821

M3-NAFLD aims to improve the clinical management of non-alcoholic fatty liver disease (NAFLD) - the leading cause of liver mortality and morbidity – by exploring the role of the gut microbiome in NAFLD pathogenesis and developing highly explainable machine learning (ML) models for superior and non-invasive screening and staging of NAFLD patients. NAFLD is a chronic disease defined by the accumulation of fat in liver cells in the absence of excessive alcohol consumption. Driven by the increasing prevalence of obesity and type 2 diabetes, NAFLD threatens more than 32% of the world population, with advanced stages such as fibrosis, cirrhosis, and hepatocellular carcinoma becoming ever more common. Despite its alarming prevalence, the clinical management of NAFLD is insufficient to respond to its escalating societal and economic burden. Critical limitations at the level of diagnosis (reliance on liver biopsy) and staging (complex pathogenesis associated with multiple metabolic factors), as well as limited awareness of the disease in the medical community, result in most early stages being neglected until late and potentially irreversible stages of the disease. In M3-NAFLD, and thanks to our well-phenotyped cohort for studying NAFLD with access to conventional (clinical and biochemical) and unconventional (omics and meta-omics) data, we have set a plan to: • characterise microbiome communities and host metabolites and lipids at different histologically proven stages of the disease to expand the current understanding of NAFLD pathogenesis, • develop highly explainable ML models for superior and histology-free screening and staging of NAFLD patients, • explore ML models for phenotyping at the primary (population-wide and low-resolution screening) and secondary (personalised and high-resolution staging) level that can support a paradigm shift from a liver-centred histology-dependant to a patient-centred, non-invasive multi-type biomarkers definition of the disease. Integration of this vast amount of data will shed light on the intricate molecular interactions between the gut microbiome, the host metabolome, and the disease pathogenesis, with emphasis on the two prominent NAFLD landmarks, hepatic inflammation and fibrosis, and this may serve as a starting point for the development of microbiome-based therapeutics. Finally, by exploring different families of ML methods, we expect to produce highly explainable, accurate models that can be interpreted by clinicians, moving away from impossible-to-argue black-box ML models toward a pathway leading to medical applicability.

• DFG Programme: Research Grants