Multi-region transcriptomics and unsupervised clustering reveal recurrent and region-specific gene signatures in Alzheimer's disease

Alzheimer's disease (AD) exhibits anatomically heterogeneous transcriptomic alterations, complicating the identification of disease-associated genes with consistent external support. We compared four gene-prioritization strategies, absolute log₂ fold change, false discovery rate (FDR), CombinedScore, and RankSum, using laser-capture-microdissected neuronal expression profiles from six brain regions in GSE5281. Differential expression was assessed separately within each region using limma, followed by within-region gene prioritization and cross-region comparison to identify recurrent and anatomically specific candidates. Robustness was evaluated through resampling stability, sensitivity to candidate-list size and scoring formulation, and external benchmarking in three anatomically matched regions of GSE48350. STAT4, SST, and NVL emerged as recurrent candidates, whereas ZNRD2, ANKRD12, and NME7 were prominent regionally prioritized signals. External directional agreement across Top-100 lists ranged from 68.3% to 72.0%; RankSum achieved 71.7% agreement and the highest or joint-highest directional agreement at Top-50 and Top-200. A fixed expression score comprising the three recurrent genes showed higher values in AD in the external cohort. Exploratory six-gene clustering revealed diagnosis-associated structure with modest separation. 14 of 18 gene-region comparisons showed concordant AD-control directions, although cross-dataset effect-size correlations were weak. This study establishes a comparative framework linking regional gene prioritization with stability assessment and independent-cohort benchmarking, identifying recurrent AD-associated candidates and quantifying their external support.