Principal component analysis (PCA) is an unsupervised method of pattern recognition. PCA can reveal visually the natural grouping of samples (Xu et al., 2017 (link); Hu et al., 2020 (link)). Each point represents a sample in a PCA score plot, which is convenient for the removal of outliers (Zhou et al., 2017 (link)). Whereas, orthogonal projections to latent structures discriminant analysis (OPLS-DA) is carried out for supervised regression modeling and identifies potential differential biomarkers between groups (Wang, T. et al., 2019 (link)). The values of R2Y and Q2 are key indices to evaluate the fitting quality and predictability of OPLS-DA models. The values of Q2 were larger than 0.5 and the difference between R2Y and Q2 was less than 0.3, suggesting superior quality of our models. (Wang, X. et al., 2020 (link)).
Variable importance in projection (VIP) is implemented to measure the contribution to sample classification (Yuan, Z. et al., 2020 (link)). Coupled with VIP values (VIP >3) and the Student’s t-test (p < 0.05), the m/z of significantly altered metabolites was filtered preliminarily. According to the retention time, an accurate molecular weight was determined through DataAnalysis 4.4 software (Bruker). And next, biomarker (error <5 ppm) were identified through the Human Metabolome Database (