Fig. 4.

Lipids and metabolites’ classification of patients with NAFLD before and after the metabolic tipping point of NAFLD.

Results of random forest predicting (A and B) NAFLD Kleiner fibrosis score (0–1) vs. (2–4), and (C and D) Kleiner fibrosis score (0–2) vs. (3–4). (A) and (C) show the progress of recursive feature addition, recursively adding the most important features, plotting median AUC of the model as more features are added. (B) and (D) show the ROC curves (main panel) and distribution of AUCs (inset box plot) when using the minimal feature sets (n = 2,001 iterations). LM, lipid and metabolite; LPC, lysophosphatidylcholine; NAFLD, non-alcoholic fatty liver disease; PC, phosphatidylcholine; PE, phosphatidylethanolamine; SM, sphingomyelin; TG, triacylglycerol.