Using Machine learning to identify metabolomic signatures of ulcerative colitis with the classification according to disease extent

Abstract Background: Ulcerative colitis (UC) is a metabolic-related chronic intestinal inflammatory disease. Disease extent is an intrinsic aspect of UC and influences the treatment modality. Identifying serum metabolic profiling of different disease extent may help to elucidate the underlying molecular mechanisms of UC. Methods: UC patients with different disease extent and healthy controls were included in this study. Liquid chromatography-tandem mass spectrometry (LC-MS/MS) method detected the metabolites, and the orthogonal partial least squares-discriminant analysis (OPLS-DA) were performed to identify metabolites. Lasso regression and support vector machine-recursive feature elimination (SVM-RFE) were used to screen metabolites. Receiver operating characteristic curves (ROC) were plotted to represent the diagnostic value of metabolites. Results: 93 differential metabolites between the UC and HCs, and57, 66, 44 in E1 and E2, E1 and E3, E2 and E3 groups were confirmed by OPLS-DA in the positive and negative ion mode. KEGG pathway analysis show that Arachidonic acid metabolism, Glycerophospholipid metabolism and Pentose phosphate pathway were the main disturbed metabolic pathway related to UC. Glycerophospholipid metabolism, Sphingolipid metabolism, Pyrimidine metabolism and Linoleic acid metabolism are involved UC site phenotype. Lasso and SVM-RFE screened differential metabolites relating to disease extent, and 8 metabolites were confirmed. The area under the ROC curve of all metabolites predicting the disease extent is greater than 0.7. Conclusions: There are significant differences in metabolomics among UC patients with different disease extent. Tridecanoic acid may be a potential biomarker for identifying patients with extensive colitis.