Identification of hub genes involved in the pathogenesis of ulcerative colitis: a multi-omics study integrating machine learning, Mendelian randomization, and mediation analysis

Background Ulcerative colitis (UC) is a chronic inflammatory bowel disease with a steadily increasing global prevalence, yet its precise molecular mechanisms remain incompletely understood. This study aimed to identify potential diagnostic biomarkers for UC and explore potential metabolite-mediated pathways linking gene expression to UC. Methods Gene expression datasets were obtained from the Gene Expression Omnibus database and analyzed using differential expression analysis, weighted gene co-expression network analysis (WGCNA), protein–protein interaction (PPI) network analysis, and multiple machine learning (ML) algorithms to identify key UC-associated genes. Mendelian randomization (MR) analysis based on cis-expression quantitative trait loci was performed to assess potential causal associations between candidate gene expression and UC, with the random-effects inverse-variance-weighted method used as the primary analytical approach. Exploratory two-step MR mediation analysis was subsequently conducted to investigate potential metabolite-mediated pathways. In addition, core-gene expression was assessed in a dextran sulfate sodium-induced mouse model of colitis. Results By integrating transcriptomic analysis, WGCNA, PPI network analysis, and 127 ML model combinations, we identified six potential diagnostic biomarkers for UC: CXCL1, CXCL2, CXCR2, ENPP3, MMP1, and PPBP. The primary MR analyses showed inverse associations of CXCR2 and ENPP3 expression with UC risk (CXCR2: OR = 0.5792, 95% CI: 0.4596–0.7299; P = 3.69 × 10 −6 ; ENPP3: OR = 0.3111, 95% CI: 0.1987–0.4871; P = 3.33 × 10 −7 ; both BH-adjusted P < 0.001). Exploratory two-step MR mediation analysis suggested that arachidonoylcholine may partially mediate the association between CXCR2 expression and UC (indirect effect: β = −0.0566, 95% CI: −0.1041 to −0.0092; P = 0.0193), accounting for an estimated 10.37% of the total effect. Conclusion This study identified six UC-associated candidate biomarkers and suggested that plasma arachidonoylcholine may be a candidate mediator of the inverse association between whole-blood CXCR2 expression and UC risk. These findings require further validation through functional experiments.