Investigation of Stress Granule-Associated Biomarkers (IGF1 and CDK2) and Their Regulatory Mechanisms in Alzheimer’s Disease: A Combined Transcriptomic, Bioinformatic and Experimental Study

Background Alzheimer's disease (AD) is the leading neurodegenerative disorder in older adults. Stress granules (SGs) have increasingly been implicated in AD, but their regulatory circuitry and biomarker relevance remain insufficiently defined. Methods Public transcriptomic datasets were integrated and analyzed with machine-learning strategies. Candidate biomarkers were refined by expression validation and receiver operating characteristic (ROC) analysis. A fully connected neural network (FCNN) was constructed to estimate diagnostic performance for AD. Gene set enrichment analysis (GSEA), molecular-network assessment, and immune-infiltration profiling were then performed to characterize biomarker-related functions. Reverse transcription quantitative polymerase chain reaction (RT-qPCR) was used to verify biomarker expression in clinical samples. Results Machine learning yielded seven feature genes, from which IGF1 and CDK2 were retained as core biomarkers for downstream analyses (AUC > 0.7). Across datasets, IGF1 was consistently decreased, whereas CDK2 was elevated in AD (p < 0.05). The FCNN achieved an AUC of 0.989, supporting strong diagnostic discrimination. Regulatory-network analysis identified miRNA associations involving IGF1-hsa-miR-19b-3p and CDK2-hsa-miR-302a-3p, and both biomarkers were enriched in the KEGG T-cell receptor signaling pathway. Molecular docking showed binding affinities of -7.3 kcal/mol for IGF1-N,N-bis(3-(D-gluconamido) propyl) deoxycholamide and − 11.2 kcal/mol for CDK2-Alvocidib. RT-qPCR confirmed IGF1 downregulation in AD, whereas CDK2 differential expression was not verified, which may reflect peripheral-blood detection limitations or cohort-specific characteristics. Conclusion Collectively, this study identified IGF1 and CDK2 as potential AD‑related biomarkers via machine‑learning analysis of public transcriptomic data. The FCNN model exhibited good diagnostic performance, and relevant miRNA regulatory axes and candidate small‑molecule compounds were predicted. Although IGF1 expression alteration was validated by RT‑qPCR in clinical samples, CDK2 differential expression was not replicated, possibly due to peripheral blood detection constraints or cohort heterogeneity. This work provided preliminary clues for AD diagnostic markers and therapeutic targets, and further validations in larger cohorts are required.