Intestinal transcriptomic analysis and prediction of biomarkers associated with mucosal healing following vedolizumab treatment in ulcerative colitis using machine learning

BACKGROUND: There is an increasing need for predictors of clinical outcomes in patients with ulcerative colitis (UC). In this exploratory analysis, we aimed to analyze the gene expression profile of intestinal tissues according to treatment outcome in patients with UC treated with vedolizumab and to develop a treatment outcome prediction model. METHODS: Transcriptomic analysis of intestinal tissues at the inflammation site (week 0) of patients (N = 10), stratified by mucosal healing status at week 54 after vedolizumab 300 mg treatment, was performed to identify differentially expressed genes (DEGs) with ≥ 1.5-fold difference and P < 0.05. Vedolizumab-related ‘hub’ genes were identified by network analysis (cytoHubba). Candidate biomarkers were selected and a prediction model for mucosal healing was developed using the least absolute shrinkage and selection operator (LASSO) algorithm (a machine learning method that helps to identify the most relevant genes for prediction while avoiding overfitting), which was validated with a public dataset (GSE73661). RESULTS: Transcriptomic analysis revealed 375 DEGs associated with vedolizumab mucosal healing. Gene enrichment analysis revealed terms associated with T cell activation and immune regulation. Eighteen hub genes were identified. A prediction model developed and validated with a public dataset had AUC–ROC values of 0.800 (95% CI, 0.551–1.000) and 0.750 (95% CI, 0.350–1.000) for the week 12 and 52 vedolizumab validation cohorts, respectively. In contrast, the AUC–ROC was 0.575 (95% CI, 0.317–0.833) for the week 4–6 infliximab validation cohort, suggesting that the model is specific for vedolizumab-associated clinical outcomes. CONCLUSIONS: We identified hub genes that were related to vedolizumab treatment outcomes. A prediction model of vedolizumab-associated mucosal healing was developed and validated with a public dataset.