Spatial Transcriptomic Deep Learning and Gene-Based Machine Learning for SLE Diagnosis: A Comparative Analysis

Systemic Lupus Erythematosus (SLE) is a chronic autoimmune disorder marked by multi-organ manifestations, high clinical variability, and complex immune dysregulation, particularly through interferon-related pathways. Early detection and effective patient stratification remain major challenges due to the absence of reliable molecular biomarkers. In this study, we analyzed gene expression data from the publicly available GSE50772 dataset, which contains microarray profiles of peripheral blood mononuclear cells (PBMCs) from SLE patients and healthy individuals. Probe-level data were mapped to gene symbols using Affymetrix CDF and GPL570 annotations, aggregated at the gene level, and subjected to feature selection with ANOVA F-test and mutual information. Dimensionality reduction with principal component analysis (PCA) was employed before training multiple machine learning classifiers, including logistic regression, support vector machine (SVM), random forest, and XGBoost. Deep learning approaches, such as two-dimensional convolutional neural networks (CNNs), achieved an accuracy of 80% with perfect recall, while classical models like logistic regression and SVM reached 100% accuracy and AUC scores at the gene level. These findings highlight the promise of integrating transcriptomic data with computational models to improve biomarker discovery and support diagnostic strategies in autoimmune diseases such as SLE.