Fig. 8
From: Predicting diabetic kidney disease with serum metabolomics and gut microbiota
Workflow chart of feature selection. For a total of 180 metabolites, univariate analyses (Wilcoxon’s rank-sum test) and OPLS-DA were employed for three clinical aims (aim 1: DKD vs. NC, aim 2: DKD vs. DM, aim 3: DKD middle vs. DKD early). Ten metabolites with p < 0.05 and VIP > 1 in all three clinical aims were selected and fed into least absolute shrinkage and selection operator (LASSO) for three aims. The overlap of top 3 LASSO non-zero coefficients was selected to yield the final panel three metabolites. “AND” means the intersection of two or more sets.
