Classification of metabolites
The obtained specimens were thawed slowly at 4°C, extracts were added and centrifuged several times, and the supernatants were separated and detected by LC-MS/MS technology for metabolite separation and detection. The data were processed by Compound Discoverer 3.1 (Thermo Fisher Scientific, USA) (mainly including peak extraction, retention time correction within and between groups, combined ion merging, missing value filling, background peak labeling, and data quality control), and then the molecular weight, retention time, peak area, and identification results were combined with KEGG (Kyoto Gene and Kyoto Encyclopedia), BGI Library, Chemspider database, Lipidmaps database, and HMDB database for identification, taxonomic annotation, and pathway annotation of the obtained metabolites to obtain the classification of metabolites in negative ion mode (Figure 2a) and the metabolic pathways they are involved in (Figure 2b).
Screening for differential metabolites
In the two groups of samples, after normalizing the data in both positive and negative ion modes, a total of 7459 metabolites were detected, which were validated by performing principal component analysis (PCA, Figure 3a) on the metabolites while applying orthogonal partial least squares-discriminant analysis (OPLS-DA, Figure 3b), which was validated by OPLS-DA. The differences between the two groups of samples were quite significant, and the samples were basically in the 95% confidence interval (CI). We were also able to obtain the value of the variable impact importance factor (VIP) of the first component in OPLS-DA. This summarizes the contribution of each variable to the model, and we considered metabolites with VIP > 1 and p < 0.05 as metabolites with significant differences, where the VIP values of the first 50 metabolites are shown in Figure 3c. According to the volcano plot (Figure 4a) , it can be concluded that the sepsis group had 370 high metabolites and 386 low metabolites compared to the normal group, with the top 25 differential metabolites visualized in the heatmap (Figure 4b). The differential metabolites obtained by the above analysis often have similar and complementary functions biological, or are positively or negatively regulated by the same metabolic pathway and show similar or opposite expression characteristics between the sepsis and normal groups, and we applied MetaboAnalyst 5.0 software to classify the above metabolites into the organic heterocyclic compound class, organic acid class, organic carbohydrates, organic hydroxides, nucleic acids, benzenes, sterol lipids, organohalogen compounds, isoprenoids, polyketides, and fatty acyl groups.
Metabolomics pathway analysis
The metabolites were analyzed by MetaboAnalyst 5.0 online software for metabolic pathways, and the general overview of metabolic pathways is shown in Figure 5, while potential differential metabolic pathways were screened by combining the KEGG database with effect values >0.5 and P values <0.05. A total of four differential metabolic pathways were screened, namely, caffeine metabolism,biosynthesis of phenylalanine, tyrosine and tryptophan. linolenic acid metabolism and phenylalanine metabolism in Supplementary Table S2.. Combined with the previous volcano and heatmaps, the above four metabolic pathways were screened again for differential metabolites, and a total of 12 differential compounds were screened, namely, 3-phenyl lactate, N-phenylacetylglutamine, phenylethylamine, traumatin, xanthine, methyl jasmonate, indole, levotryptophan, 1107116, traumatic acid, theobromine, and salicylic acid.
The diagnostic value of different metabolites in sepsis
To determine the diagnostic efficiency of the above 12 potentially differential metabolites, we generated subject operating characteristic curves (ROC curves) and calculated the area under the curve (AUC) for each incorporated characteristic, according to the Swets judgment criteria [24]. AUC < 0.5 indicates that the test has no diagnostic value, AUC of 0.8-0.9 indicates that the test has good accuracy, and AUC > 0.9 indicates that the diagnostic test has high accuracy. Finally, we obtained nine differential metabolites whose expression values were significantly different between sepsis and normal control species, namely, 3-phenyl lactate (AUC: 0.923), N-phenylacetylglutamine (AUC: 0.782), phenylethylamine (AUC: 0.825), traumatin (AUC: 0.941), xanthine (AUC: 0.900), methyl jasmonate (AUC: 0.823), indole (AUC: 0.909), levotryptophan (AUC: 0.859), and 1107116 (AUC: 0.916), whose ROC curves are shown in Figure 6.