Fig. 7: Strength of association of selected predictors for features of NAFLD/NASH and liver fat in patients with NAFLD/NASH.
From: BMP4 and Gremlin 1 regulate hepatic cell senescence during clinical progression of NAFLD/NASH
a–f, Relative feature importance for NAFLD/NASH features (including steatosis, ballooning, inflammation and fibrosis score) and liver fat using predictive machine learning models such as conditional random forest, gradient boosting models and partial dependence plots. Predictors that display a pronounced increase in relative importance are considered strong predictors for the outcome. Model diagnostics (that is, R2 and r.m.s.e.) for predictive machine learning models are presented in each panel. Partial dependence plots were used to investigate interaction effects between important features. To assess significance level and estimate risk association for important features, according to machine learning models, we subsequently constructed either a linear or logistic regression model and included the most important features. AT, adipose tissue.
