Machine learning and metabolomics identify biomarkers associated with the disease extent of ulcerative colitis

Ulcerative colitis (UC) is a metabolism-related chronic intestinal inflammatory disease. Disease extent is a key parameter of UC. Using serum metabolic profiling to identify noninvasive biomarkers of disease extent may inform therapeutic decisions and risk stratification.

The orthogonal partial least squares-discriminant analysis (OPLS-DA) was performed to identify the metabolites. Least absolute shrinkage and selection operator regression, random forest-recursive feature elimination, and support vector machine-recursive feature elimination algorithms were used to screen metabolites. Five machine learning algorithms (eXtreme Gradient Boosting, K-NearestNeighbor, Naive Bayes, random forest [RF], and SVM) were used to construct the prediction model.

A total of 220 differential metabolites between the patients with UC and healthy controls (HCs) were confirmed by the OPLS-DA model. Machine learning screened 8 essential metabolites for distinguishing patients with UC from HCs. A total of 23, 6, and 6 differential metabolites were obtained through machine learning between groups E1 and E2, E1 and E3, and E2 and E3. The RF model had a prediction accuracy of up to 100% in all 3 training sets. The serum levels of tridecanoic acid were significantly lower, and pelargonic acid was significantly higher in patients with extensive colitis than in the other groups. The serum level of asparaginyl valine in patients with rectal UC was significantly lower than that in the E2 and E3 groups.

Our findings revealed the metabolic landscape of UC and identified biomarkers for different disease extents, confirming the value of metabolites in predicting the occurrence and progression of UC.