Identification of 5 Potential Predictive Biomarkers for Alzheimer's Disease by Integrating the Unified Test for Molecular Signatures and Weighted Gene Coexpression Network Analysis

Abstract

Background: Previous transcriptome-wide association study (TWAS) has documented 21 genes associated with Alzheimer's disease (AD) risk, but the predictive biomarkers remain unexplored.

Methods: TWAS leveraging the unified test for molecular signatures (UTMOST) was performed in 75,000 cases and 420,000 controls with 10 brain tissue gene expression references. Weighted gene coexpression network analysis (WGCNA) was conducted in GSE5281 and GSE48350 data sets containing 167 AD samples and 247 controls. Random forest (RF) analysis was applied to screen the potential predictive biomarkers based on overlapping genes identified by TWAS and WGCNA, followed by comprehensive bioinformatic analyses with differential gene expression, functional enrichment, and correlation with immune cells. A nomogram was established to verify the predictive power of the identified biomarkers.

Results: TWAS revealed 78 candidate genes (p < 2.89 × 10-6). In WGCNA turquoise module, 3 718 AD-related genes were screened. RF identified 5 predictive biomarkers (FAM71E1, DDB2, AP4M1, GPR4, DOC2A), which are enriched in the global genome nucleotide excision repair pathway and associated with immune cell designations "Natural.killer.T.cell," "Memory.B.cell," "T.follicular.helper.cell," "Neutrophil," and "MDSC." The nomogram based on the 5 markers showed a high predictive power.

Conclusion: Five potential predictive biomarkers for AD were identified, providing new insights into the pathogenesis and etiology of AD.

Keywords: AD; Random forests; TWAS; UTMOST; WGCNA.