Abstract
Alzheimer’s disease (AD) is a multifactorial neurodegenerative disorder characterized by complex molecular alterations across multiple brain regions. In this study, we applied an integrative systems-level framework combining multi-region transcriptomic analysis, protein–protein interaction (PPI) network topology, machine learning–based validation, and in silico approaches to identify robust and pharmacologically relevant molecular targets in AD. Gene expression data from four AD-related brain regions (entorhinal cortex, frontal cortex, hippocampus, and temporal cortex) were obtained from the GSE5281 dataset, yielding 467 genes consistently dysregulated across all regions. Network analysis identified a subset of downregulated hub genes predominantly associated with mitochondrial bioenergetics and proteostasis pathways. Intersection with experimentally curated targets of genistein and resveratrol prioritized four elite genes, COX5B, ENO1, HSP90AB1, and SDHB as shared and biologically central nodes. To assess the collective discriminative capacity of the identified gene set, a random forest classifier was trained using shared differentially expressed genes. The model demonstrated strong classification performance with a low out-of-bag error rate, while feature importance analysis identified SDHB as the most influential contributor, followed by COX5B and ENO1, indicating synergistic multigene effects rather than isolated biomarker behavior. Machine learning was applied as an integrative validation layer to reinforce biological relevance, rather than as a standalone diagnostic tool. Finally, molecular docking analyses revealed favorable binding affinities of genistein and resveratrol toward all four elite targets. Overall, these findings highlight a convergent mitochondrial–proteostasis dysfunction axis in AD and suggest COX5B, ENO1, HSP90AB1, and SDHB as promising multi-target nodes for polyphenol-based therapeutic strategies.