Untargeted metabolic profiles discriminate between different inflammatory asthma phenotypes and healthy subjects in the discovery set
The score plot of OPLS-DA analysis showed a distinct separation between different asthma phenotypes and the Asthma group was also clearly separated from the Healthy group (Figure 1). The parameters for the classification models from the software were showed in the Table E3, suggesting all of the model had a very good fitness and predictive capability. Furthermore, a leave-one-out cross-validation (LOOCV) was used to estimate the robustness and predictive ability of each model, and thus a 200 permutation test was applied. The low value of the Q2 intercept in each model indicated the robustness of these model and thus showed a low risk of over-fitting in all of the comparison.
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, a total of 77 differential metabolites were screened between different asthma phenotypes and healthy subjects as shown in the heatmap in the Figure 2.