Univariate and multivariate analysis for metabolomic profiling
After high-throughput screening of the metabolic features using the above rules, a total of 772 (473 from positive ion mode and 299 from negative ion mode). For each metabolite peak reproducibly detected in whole samples, the unpaired two-tailed Wilcoxon rank-sum test (a nonparametric univariate method) was carried out to measure the significance of each metabolite among different groups and adjusted by false discovery rate (FDR) correction (Benjamini–Hochberg method). The merged UHPLC-MS data were log-transformed for subsequent analyses. The filtered and normalized data matrix was exported for multivariate statistical analysis using principal component analysis (PCA) and orthogonal partial least-squares discriminant analysis (OPLS-DA) with SIMCA 16.1 software package (Umetrics, Umea, Sweden). To further validate the model, the permutation tests (n = 200) was proceeded. By combining the univariate and multivariate statistical analysis, on the basis of a variable importance in the projection (VIP) threshold of 1 from the 10-fold cross-validated OPLS-DA model, fold change < 0.83 or > 1.2, FDR < 0.05, numbers of different metabolites among different groups were obtained.