Integrative Transcriptomics and Machine Learning Reveal a Robust Biomarker Signature for Alzheimer’s Disease

Background: Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline and complex molecular alterations involving immune activation, synaptic dysfunction, and metabolic imbalance. Despite extensive research, robust and biologically interpretable transcriptomic signatures capable of reliably distinguishing AD from cognitively normal controls remain limited. This study aimed to identify a stable gene expression signature associated with AD and evaluate its predictive performance using complementary machine learning approaches. Methods: Publicly available transcriptomic datasets were analyzed to identify differentially expressed genes between AD patients and controls. Functional enrichment analysis was performed to characterize the biological pathways associated with the identified genes. Multiple supervised machine learning algorithms were implemented to evaluate classification performance, and cross-model feature importance analysis was used to refine a stable and discriminative gene set. The final optimized feature subset was evaluated using an XGBoost classifier to assess predictive robustness. Results: Differential expression analysis revealed significant dysregulation of genes involved in neuroinflammatory processes, immune signaling pathways, synaptic transmission, and metabolic regulation. Functional enrichment confirmed the predominance of immune activation and neuronal dysfunction pathways. Machine learning models consistently captured these transcriptomic alterations, demonstrating strong classification performance with substantial agreement across algorithms. Feature refinement led to the identification of a stable gene signature that achieved high predictive accuracy within the XGBoost framework, supporting both its statistical robustness and biological relevance. Conclusion: The integration of transcriptomic profiling with complementary machine learning strategies enabled the identification of a reproducible and biologically coherent molecular signature of Alzheimer’s disease. These findings highlight the utility of combining differential expression analysis with predictive modeling to uncover disease-relevant molecular patterns and support the potential application of the identified gene signature in biomarker development and molecular stratification of AD.