Fig. 4

Establishment of integrated biomarker profiling. (A) AUC of the integrated 412 differential metabolites based on random forest classification (RFC) model. (B) the Mean Decrease in Accuracy (MDA) of 20 potential biomarkers. (C) Gini impurity of 20 potential biomarkers. (D) Distribution of 5 trials of 10-fold cross-validation error in random forest classifiers. The model was trained with 412 differential metabolites in the training set (T-T2DM group, n = 73; diabetes group, n = 67). The black solid curve showed the trials. The red line indicated the number of picked features in the optimal set. (E) AUC of the 5 selected potential biomarkers from the RFC model. (F) The prediction performance of the model consisted of 5 potential biomarkers in the train and test sets. (G) ROC curves for traditional markers BMI, fasting glucose and HbA1c. (H) ROC curves for risk factors of T-T2DM (age, triglycerides, serum uric acid and creatinine)