Identification and Experimental Validation of Oxidative Stress-Related Biomarkers in Ulcerative Colitis Using Machine Learning.

Purpose: Ulcerative colitis (UC) remains challenging to diagnose and treat due to a lack of reliable biomarkers. This study investigates oxidative stress-related targets in UC using bioinformatics and experimental validation. Methods: We analyzed four GEO datasets and oxidative-stress genes from MSigDB, applying differential analysis, LASSO regression (for feature selection), and random forest (for robust biomarker identification). An artificial neural network (ANN) diagnostic model was constructed, followed by chromosomal distribution analysis, immune infiltration assessment, and drug screening. Hub gene expression was validated in a 3% DSS-induced colitis mouse model via qPCR and Western blot. Results: Ultimately there were 6 hub genes identified: DUOX2, ETFDH, GPX8, ITGA5, NPY, and PDK2, which were validated with 3 other datasets. In the DSS-colitis model, DUOX2 and ITGA5 were significantly upregulated (p < 0.05), whereas ETFDH, PDK2, and NPY were downregulated. GPX8 protein expression was elevated in colonic mucosa compared to controls. These findings were further validated in three independent datasets (GSE48958, GSE16879, GSE36807). Conclusion: Our study identifies six oxidative stress-related biomarkers in UC using machine learning and experimental validation. These findings provide potential diagnostic and therapeutic targets for UC management, paving the way for further clinical investigations.