Prediction of inflammatory bowel disease recurrence risk based on multi-omics integration and machine learning models.

PURPOSE: Postoperative recurrence remains a major clinical challenge in inflammatory bowel disease (IBD), while robust molecular predictors with clear biological and causal relevance are still lacking. This study aimed to identify and validate recurrence-associated genes in IBD using an integrative multi-level analytical framework. METHODS: Two independent postoperative IBD transcriptomic datasets were analyzed to identify recurrence-related differentially expressed genes and co-expression modules. Mendelian randomization (MR) and colocalization analyses were subsequently applied to prioritize genes with potential causal relevance. Key genes were validated by quantitative real-time PCR (qRT-PCR) in independent clinical ileal tissue samples. Machine learning models were constructed and externally validated in a real-world clinical cohort to assess recurrence prediction performance. RESULTS: Integrative analyses identified 253 recurrence-related genes, including 73 shared upregulated genes, 31 shared downregulated genes, and 157 shared module genes. MR analysis identified six candidate genes, among which KLHL24, NPC1, and PSMG1 showed strong colocalization evidence (PP.H4 > 0.95). qRT-PCR validation confirmed that KLHL24 and NPC1 were significantly downregulated, whereas PSMG1 was significantly upregulated in recurrent ileal tissues compared with non-recurrent tissues (all P < 0.001). In the external validation cohort of 178 patients, the logistic regression model showed the highest predictive performance, with an AUC of 0.801 (95% CI: 0.781-0.820), sensitivity of 0.712, and specificity of 0.741, followed by random forest, XGBoost, and decision tree models with AUCs of 0.792, 0.787, and 0.750, respectively. CONCLUSION: This study delineates molecular features associated with postoperative recurrence in IBD and identifies key genes supported by transcriptomic, genetic, and clinical evidence, providing a robust basis for recurrence risk stratification.