To identify the differential metabo

To identify the differential metabolites of CRC, the metabolomesof tumor tissues were compared with that of matched adjacentmucosa. Supplementary Table 1 shows the demographiccharacteristics and clinicopathological features of 51 CRCpatients. Mass spectrometry detected 4,526 and 4,765 variables innegative electrospray ionization (ESI-) and positive electrosprayionization (ESI+), respectively. Multivariate analysis wasperformed on the result of mass spectrometry to find metabolitesthat mostly discriminated the study groups. Principal componentanalysis (PCA) was the unsupervised analysis method, whichwas used for dimension reduction of data through making alinear combination of variables known as principal components.PCA analysis can reveal trends in the data and groups ofobservations and find outliers. Although a weak trend inclustering according to the PCA plot based on tumor tissueand adjacent mucosa was observed, the PCA analysis resultsshowed a separation of tumor tissue and adjacent mucosa intotwo clusters (Figures 1A,D). To further study the differencesbetween tumor tissue and adjacent mucosa and to find potentialbiomarkers, the supervised multivariate statistical method OPLSDA was subsequently used. OPLS-DA is a supervised analysismethod that is employed to divide the samples into differentgroups, including tumor tissue and adjacent mucosa, which wasperformed to find metabolites that mostly discriminated thestudied groups in each comparison. The classification resultsare shown in Figures 1B,E. To guard against model overfitting,permutation tests (100 random permutations) were performed.These permutation tests were used to contrast the goodness offit of the original model with the goodness of fit of randomlypermuted models. As shown in Figures 1C,F, the validationplots strongly indicated that the original combined models werevalid. No overfitting was observed.A total of 373 metabolites (296 higher and 77 lower) wereidentified with the criteria of Variable important for theprojection (VIP) score >1.5 and P-values of