Additional file 1: Figure S1. of Metabolomic biosignature differentiates melancholic depressive patients from healthy controls

Effect of storage time correction. Figure S2. The distributions of original and imputed metabolite features: Glyoxylate ratio, Caffeine ratio, Elaidicacid ratio and Indole 3 propionic acid ratio. Figure S3. QQ plots of the p-values of the two-sample t-tests on raw features, k-means and hierarchical clustering representatives. Table S1. Classification performance obtained by Random Forest on metabolite data using the standard undersampling technique. Table S2. Classification performance obtained by Support Vector Machines on metabolite data using the standard undersampling technique. Table S3. Top 30 individual metabolic features selected by different feature selection methods. Table S4. Top 30 individual metabolic features selected by different feature selection methods. Table S5. Top cluster-representatives (K-means) selected by different feature selection methods. Table S6. Top cluster-representatives (hierarchical clustering) selected by different feature selection methods. Table S7. Top cluster-representatives (hierarchical clustering) selected by different feature selection methods. (DOCX 812Â kb)