Abstract <p>Alzheimer’s disease (AD) is a complex neurodegenerative disorder characterized by widespread dysregulation of gene expression and regulatory pathways. MicroRNAs (miRNAs) act as key post-transcriptional regulators by modulating messenger RNAs (mRNAs), and their disruption can influence synaptic function, neuroinflammation, and neuronal survival. In this study, we present a transcriptomic-driven framework in which differentially expressed genes (DEGs) are identified from gene expression data and integrated with curated miRNA–target interaction databases to infer putative AD-associated miRNA–mRNA regulatory signatures and potential candidate biomarkers. Transcriptomic and clinical data were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), and the GEO dataset GSE48552 was used as supplementary support to assess the consistency of observed transcriptomic patterns. Using an exploratory differential expression threshold with Welch’s t-test and FDR correction, 123 candidate dysregulated genes (34 up-regulated, 89 down-regulated) were identified between AD and cognitively normal controls. To further assess robustness, threshold-sensitivity and cross-method concordance analyses were conducted, supporting the presence of a reproducible core transcriptional signal within the broader discovery-level DEG set. Experimentally validated and predicted miRNA–target interactions were integrated using miRTarBase, yielding 1,669,089 miRNA–gene interactions involving 3,055 unique miRNAs, with strong enrichment toward down-regulated gene targeting. Functional enrichment analysis revealed convergence of miRNA-regulated genes on synaptic signaling, neuronal communication, intracellular transport, apoptosis, oxidative stress, and PI3K–Akt/MAPK-related pathways. A bipartite putative miRNA–mRNA regulatory network (2,207 nodes connected by 11,437 edges, including 2,104 miRNAs and 103 significant genes) was constructed and analyzed using centrality metrics, prioritizing candidate hub genes, including PBX1 and KREMEN1, which were subsequently interpreted in the context of neuronal transcriptional regulation, Wnt-related signalling, synaptic vulnerability, and AD-associated pathway enrichment. Finally, supervised machine learning models trained on selected molecular features showed discriminative performance in the held-out test set, with Random Forest, Gradient Boosting, and LightGBM achieving the highest ROC–AUC values, indicating strong capability in distinguishing AD from control samples. Overall, the framework provides a biologically interpretable strategy for biomarker discovery, prioritizing AD-associated candidate biomarkers and putative regulatory interactions while highlighting targets for future experimental and clinical validation.</p> Graphical Abstract <p></p>