Non-alcoholic fatty liver disease (NAFLD) is one of the metabolic diseases caused by the accumulation of fat in the liver by numerous factors such as genetic factors, bacterial imbalances, and diet, regardless of alcohol intake. NAFLD is defined as at least mild fatty liver to steatohepatitis which can cause cirrhosis and hepatocellular carcinoma. However, for most mild NAFLD patients, obvious symptoms cannot be observed except for increased levels of clinical parameters such as ALT and AST.

Recent studies have proposed decisive factor in NAFLD is dysbiosis, which causes endothelial barrier dysfunction and leaky gut. The compromised intestinal barrier allows the translocation of potential bacteria and bacteria-derived molecules to the liver through the portal vein. Since it has been shown that the gut microbiome and the liver communicate through the gut-liver axis, the gut microenvironment evaluation has been proposed for the application of sensitive and precise biomarkers. Therefore, in the current study, we applied untargeted metabolic analysis to fecal samples of healthy control and NAFLD patients using LC-orbitrap MS.

Unsupervised univariate and multivariate statistical analyses have been applied to characterize overall changes of gut microbiome-driven metabolites in NAFLD patients compared to the healthy control group. The predictive model was constructed by machine learning based on specific prioritized compounds. The diagnostic performance of the model was quantified through the Receiver Operating Characters (ROC) curve, and these results may be applied to the diagnosis and potential prognosis of NAFLD.