Integrative machine learning reveals telomere-associated gene modules reflecting neuronal dysregulation in Alzheimer's disease

Abstract

Background Telomere dysfunction contributes to cellular aging and genome instability, but telomere-associated transcription in Alzheimer's disease (AD) is poorly characterized. Telomere genes are commonly tested gene by gene, leaving it uncertain whether they organize into reproducible, disease-linked co-expression programs across cohorts. Objective To identify AD linked telomere gene modules and evaluate their diagnostic and biological relevance. Methods We curated 157 telomere maintenance genes and analyzed two GPL570-platform GEO Series datasets: GSE5281 (n = 161; AD = 87, control = 74) and GSE48350 (n = 253; AD = 80, control = 173). In GSE5281 hippocampus, RMA + age-adjusted limma identified 38 AD-associated telomere genes (FDR<0.05). Ward clustering (silhouette) defined two modules, summarized as eigengenes (module PC1). Eigengenes were evaluated across seven classifiers with nested stratified 5 × 5 cross-validation; Youden's J thresholds from out-of-fold predictions were fixed and applied to GSE48350 using frozen z-scoring (GSE5281 μ/σ) and fixed cutoffs. Robustness used 500-bootstrap stability; hub genes from the dominant module were interpreted with BRETIGEA and Reactome/GO enrichment. Results PCA and t-SNE showed AD-associated structure in telomere-gene expression. With only two module features, models achieved ROC-AUC 0.722–0.780 internally and 0.689–0.695 externally. Bootstrap resampling converged on one dominant axis: Cluster 2 was consistently the strongest feature and was reduced in AD (Cohen's d = −1.07; Welch p = 8.13 × 10 − 1 0 ). Hub genes (TSPYL5, TUBB3, PLCL2, NHP2) were downregulated, tracked neuronal signatures positively, varied inversely with microglial/astrocytic signatures, and mapped to telomere/chromosome maintenance, DNA repair, and cell-cycle regulation. Conclusions AD features a reproducible, resampling-stable telomere genome-maintenance module that provides an interpretable systems-level disease axis for mechanistic follow-up and integrative biomarker development.