A Cross-Region Meta-Analysis and Machine Learning Identifies a 37-Gene Signature Associated with Alzheimer’s Disease

Background/Objectives: Alzheimer’s disease (AD) shows marked transcriptomic heterogeneity across brain regions, limiting reproducibility. We aimed to identify robust cross-region gene signatures using meta-analysis. Methods: Differential expression (limma-voom) was performed on five bulk RNA-seq datasets (n = 230; 154 AD donors, 76 controls) from the hippocampus to cortical regions. Consensus DEGs were identified via Stouffer’s Z, random effects, and MetaVolcanoR models. Pathway enrichment and associations with Braak stage were evaluated. Validation was conducted in two independent cohorts, with predictive performance assessed using machine learning. Results: Thirty-seven consensus DEGs (16 up, 21 down; FDR ≤ 0.05) were identified across ≥4 datasets. The enrichment results revealed increased expression of glial and ECM-associated genes and decreased expression of synaptic and GABAergic genes. Thirty-six of 37 genes correlated with Braak stage, with all 37 remaining significantly associated after covariate adjustment. The signature predicted AD with AUCs of 0.784 and 0.861 for validation in two independent cohorts. Conclusions:: We identified a consistent cross-region signature linking synaptic and glial changes to neuropathological severity, highlighting new mechanisms and potential biomarkers.