Novel Alzheimer’s disease-associated variants and genetic interactions identified from UK biobank whole-exome sequencing data using IBI-DT

Abstract

Abstract Alzheimer’s disease (AD) is a highly heritable neurodegenerative disorder whose genetic architecture remains incompletely understood, particularly with respect to rare variants and higher-order interactions. We applied Individualized Bayesian Inference-Decision Tree (IBI-DT) to 790,000 whole-exome sequencing variants from 8,292 unrelated White British individuals in the UK Biobank, using genome-wide association analysis (GWAS) based on Fisher’s exact test as a marginal association comparator and evaluated the findings in an independent ADSP-Discovery ICE cohort of 1,560 unrelated individuals with 852,997 variants. In the discovery cohort, IBI-DT identified 178 significant variants using empirical null calibration, which mapped to 173 genes, whereas GWAS identified 16 significant variants that mapped to four genes. Compared with GWAS, IBI-DT prioritized more rare variants among the top 178 signals (147 vs. 66 with minor allele frequency ≤ 0.01) and identified variants with less correlated, less redundant linkage patterns. The IBI-DT-prioritized variants included established AD genes as well as additional biologically supported genes not prioritized by GWAS. Replication supported three variants and four genes across cohorts, including gene-level replication of KIF14 and ZNF90 , which were replicated only by IBI-DT. IBI-DT also identified significant gene-gene and gene-environment interactions and biologically plausible enriched pathways. Neural networks trained on IBI-DT-prioritized variants outperformed GWAS-based models (AUC 0.67 vs. 0.63). Overall, these findings indicate that IBI-DT complements GWAS by recovering established AD signals while prioritizing rare, less redundant, and interaction-related AD-associated signals.