Alterations of serum metabolic profile in major depressive disorder: A case-control study in the Chinese population

Figure 3 Receiver operating characteristic analysis revealing candidate metabolomic biomarkers for major depressive disorder diagnosis. A-E: Receiver operating characteristic (ROC) curves of 8-HETE, 10-HDoHE, cortisol, 12-HHTrE, and 10-hydroxydecanoic acid. The left panel of each picture represents the ROC for differentiating the major depressive disorder (MDD) group from the healthy control (HC) group, and the boxplots on the right side are the feature intensities of the two groups. The y-axis refers to the relative value after normalization of the peak intensities; F: Multivariate ROC curves constructed with 2-100 metabolites based on the cross-validation performance. Each curve represents the potential of the top 5, 10, 15, 25, 50, and 100 features in differentiating the MDD group from the HC group; G: Predicted class probabilities (average of the cross-validations) for each sample using the 5-feature model of metabolites; H: Corresponding predictive accuracy of each partial least squares discriminant analysis model constructed with different numbers of features. The predictive accuracy of 5 to 100 features is from 90% to 95.7%, respectively. HC: Healthy control; MDD: Major depressive disorder; AUC: Area under the receiver operating characteristic curve.