Integration of multi-omics data and machine learning to identify antioxidant biomarkers in type 1 diabetes

The identification of biomarkers for early diagnosis and monitoring the progression of Type 1 Diabetes (T1DM) is essential for improving disease management. This study integrates multi-omics data with machine learning to identify antioxidant stress proteins in serum as potential biomarkers. Serum samples from mice treated with varying doses of streptozotocin (STZ) and human transcriptomic data from the gene expression omnibus (GEO) database were analyzed using weighted gene co-expression network analysis (WGCNA). Proteomic analysis of 25 T1DM and 25 healthy controls using LC-MS/MS revealed 33 differentially expressed proteins enriched in oxidative stress pathways. Machine learning algorithms, including Random Forest and SVM-RFE, identified five key proteins: GPX3, GSTP1, PRDX6, SOD1, and MSRB2. GPX3 demonstrated the highest diagnostic value, with a significant correlation to clinical parameters such as HbA1c and fasting plasma glucose. Functional validation showed GPX3 overexpression protected pancreatic β-cells from H₂O₂-induced oxidative damage and alleviated symptoms and pathological changes in T1DM mice. These results suggest that GPX3 is a promising biomarker for diagnosing and tracking T1DM progression, offering new insights into oxidative stress management in T1DM.

Antioxidant stress proteins; Glutathione peroxidase 3; Machine learning; Multi-omics analysis; Type 1 diabetes.