Abstract
Alzheimer's disease (AD) is a neurodegenerative condition causing cognitive decline. Oxidative stress (OS) is believed to contribute to neuronal death and dysfunction in AD. We conducted a study to identify differentially expressed OS-related genes (DEOSGs) through bioinformatics analysis and experimental validation, aiming to develop a diagnostic model for AD. We analyzed the GSE33000 dataset to identify OS regulator expression profiles and create molecular clusters (C1 and C2) associated with immune cell infiltration using 310 AD samples. Cluster analysis revealed significant heterogeneity in immune infiltration. The 'WGCNA' algorithm identified cluster-specific and disease-specific differentially expressed genes (DGEs). Four machine learning models (random forest (RF), support vector machine (SVM), generalized linear model (GLM) and extreme gradient boosting (XGB)) were compared, with GLM performing the best (AUC = 0.812). Five DEOSGs (NFKBIA, PLCE1, CLIC1, SLCO4A1, TRAF3IP2) were identified based on the GLM model. AD subtype prediction accuracy was validated using nomograms and calibration curves. External datasets (GSE122063 and GSE106241) confirmed the expression levels and clinical significance of important genes. Experimental validation through RT-qPCR showed increased expression of NFKBIA, CLIC1, SLCO4A1, TRAF3IP2, and decreased expression of PLCE1 in the temporal cortex of AD mice. This study provides insights for AD research and treatment, particularly focusing on the five model-related DEOSGs.