Abstract
Background We aimed to pinpoint biomarkers, create a diagnostic model for ulcerative colitis (UC), and delve into its immune features to better understand this autoimmune condition. Methods The sequencing data for both the UC and the control groups were obtained from GEO, including both bulk and single-cell data. Using GSE87466 as training group, we applied differential analysis, WGCNA, PPI, LASSO, RF, and SVM-RFE for biomarker selection. A neural network shaped our diagnostic model, corroborated by GSE92415 as the validation cohort with ROC assessment. Immune cell profiling was conducted using CIBERSORT. Results 53 disease-associated genes were screened. Enrichment analysis highlighted roles in complement cascades and cell adhesion. Eight biomarkers were finally identified through multiple machine learning and PPI: B4GALNT2, PDZK1IP1, FAM195A, REG4, MTMR11, FLJ35024, CD55, and CD44. The diagnostic model had AUCs of 0.984 (training group) and 0.957 (validation group). UC tissues revealed heightened plasma cells, CD8 T cells, and other immune cells. Two unique UC immune patterns emerged, with certain T and NK cells central to immune modulation. Conclusion We identified eight biomarkers of UC by various methods, constructed a diagnostic model through neural networks, and explored the immune complexity of the disease, which contributes to the diagnosis and treatment of UC.