Figure legends:

Figure 1. Untargeted metabolic profiles discriminate in different inflammatory asthma phenotypes and healthy subjects in the discovery set. (A-D) Score plot of PCA model obtained from different inflammatory asthma phenotypes and healthy subjects retrospectively; (E-H) Score plot of OPLS-DA model obtained from different inflammatory asthma phenotypes and healthy subjects retrospectively; The labels t[1] and t[2] along the axes represent the scores (the first 2 partial least-squares components) of the model, which are sufficient to build a satisfactory classification model. (I-L) Permutation Test of the OPLS-DA model obtained from different inflammatory asthma phenotypes and healthy subjects retrospectively; Two hundred permutations were performed, and the resulting R² and Q² values were plotted.

Figure 2. Heatmap of identified differential metabolites in sputum samples in different inflammatory asthma phenotypes and healthy subjects in the discovery set

Figure 3. MetaboAnalyst Pathway Impact based on selected and more representative metabolites responsible for the class separation in sputum samples in different asthma phenotypes and healthy subjects. (A) Asthma and Healthy; (B) EA and NA; (C) EA and PGA; (D) NA and PGA. Circles represent metabolic pathways potentially involved in class separation. EA, eosinophilic asthma; NA, neutrophilic asthma; PGA, paucigranulocytic asthma.

Figure 4. The receiver operating characteristic (ROC) curves of metabolites to discriminate between NA and EA (A) or between NA and PGA (B). EA, eosinophilic asthma; NA, neutrophilic asthma; PGA, paucigranulocytic asthma.

Figure 5. Heatmap of correlations of differential metabolites with clinical and inflammatory profiles in all subjects with asthma in the validation set