A mucosal barrier injury related six gene transcriptomic signature associated with ulcerative colitis using machine learning and public transcriptomic data

Ulcerative colitis (UC) is characterized by epithelial barrier disruption, mucosal immune activation, and sustained intestinal inflammation. We developed a biologically interpretable public-transcriptomic signature using Gene Expression Omnibus datasets GSE75214, GSE38713, and GSE92415. A literature-curated panel of 35 mucosal barrier injury-related genes was integrated with differential expression, LASSO regression, random forest prioritization, and ridge logistic regression. Six genes (LCN2, DEFB1, S100A8, MUC1, VIL1, and TJP3) were retained as a locked transcriptomic score. Because the original feature-selection steps used the complete development cohort, we added 50 repetitions of stratified five-fold outer cross-validation. All preprocessing and ridge tuning were restricted to each outer-training fold; the locked score achieved a mean out-of-fold AUC of 0.9660 (SD 0.0031). A stricter sensitivity analysis that repeated differential-expression filtering, LASSO, random-forest ranking, deterministic six-gene selection, and ridge fitting within each outer-training fold achieved a mean out-of-fold AUC of 0.9600 (SD 0.0094). The refitted locked model retained discrimination in GSE38713 (AUC 0.8974, 95% CI 0.8040-0.9909) and GSE92415 (AUC 0.9894, 95% CI 0.9744-1.0000) when one development-derived cutoff was applied unchanged. The six-gene score did not significantly outperform VIL1 alone in the development cohort (DeLong P = 0.4026). The signature should therefore be interpreted as a molecular pattern associated with UC mucosal inflammation and barrier remodeling rather than as a UC-specific clinical diagnostic test.