To Örebro University

Background and Aims: Nonalcoholic fatty liver disease (NAFLD) is a progressive liver disease that is strongly associated with type 2 diabetes. Accurate, non-invasive diagnostic tests to deliniate the different stages: degree of steatosis, grade of nonalcoholic steatohepatitis (NASH) and stage fibrosis represent an unmet medical need. In our previous studies, we successfully identified specific serum molecular lipid signatures which associate with the amount of liver fat as well as with NASH. Here we report underlying associations between clinical data, lipidomic profiles, metabolic profiles and clinical outcomes, including downstream identification of potential biomarkers for various stages of the disease.

Method: We leverage several statistical and machine-learning approaches to analyse clinical, lipidomic and metabolomic profiles of individuals from the European Horizon 2020 project: Elucidating Pathways of Steatohepatitis (EPoS). We interrogate data on patients representing the full spectrum of NAFLD/NASH derived from the EPoS European NAFLD Registry (n = 627). We condense the EPoS lipidomic data into lipid clusters and subsequently apply non-rejection-rate-pruned partial correlation network techniques to facilitate network analysis between the datasets of lipidomic, metabolomic and clinical data. For biomarker identification, random forest ensemble classification and neural network machine learning approaches were used to both search for valid disease biomarkers and to assess the relative improvement over clinical-data-only classification versus addition of our lipidomic and metabolomic datasets.

Results: We found that steatosis grade was strongly associated with (1) an increase of triglycerides with low carbon number and double bond count as well as (2) a decrease of specific phospholipids, including lysophosphatidylcholines. In addition to the network topology as a result itself, we also present lipid clusters (LCs) of interest to the derived network of proposed interactions in our NAFLD data from the EPoS cohort, along with our proposed biomarkers for various disease outcomes, as put forward by our current machine learning analyses.

Conclusion: Our findings suggest that dysregulation of lipid metabolism in progressive stages of NAFLD is reflected in circulation and may thus hold diagnostic value as well as offer new insights about the NAFLD pathogenesis. Using this cohort as a proof-of-concept, we demonstrate current progress in tuning the accuracy of neural network and random forest approaches with a view to predicting various subtypes of NAFLD patient using a minimal set of lipidomic and metabolic markers. A detailed network-based picture emerges between lipids, polar metabolites and clinical variables. Lipidomic/metabolomic markers may provide an alternative method of NAFLD patient classification and risk stratification to guide therapy.