Uncovering Nonlinear Genetic Architectures in Alzheimer's Disease Cellular Pathogenesis - Project Summary Alzheimer's disease (AD) affects about 7.2 million Americans age 65 and older, with direct health and long-term care costs of about $384 billion in 2025. Although about 75 susceptibility loci have been mapped by genome-wide association studies (GWAS), mean-based additive analyses often miss nonlinear mechanisms, such as domi- nance, variance-modulating variants, and epistasis, especially when effects are cell-state-specific or concentrated in distribution tails. Despite 182 ongoing clinical trials testing 138 agents in 2025, attrition remains high, stressing the need for methods that link genetic risk to precise brain-cell pathways and actionable targets. This project will create a comprehensive, distribution-aware atlas of AD genetic architecture across key brain cell states. We will develop and apply quantile-focused statistical tools and multimodal analyses to identify vari- ants and interactions that shape tail-enriched molecular signatures linked to AD risk and progression. We propose the first quantile-specific Bayesian fine-mapping (qSuSiE) to produce per-quantile credible sets and posterior in- clusion probabilities; quantile-aware colocalization (qColoc) and an omnibus quantile transcriptome-wide asso- ciation study (qTWAS) to integrate evidence across quantiles, molecular layers, and cell types; an AD-focused sequence model (Evo2-AD) that generates nonlinear, cell-contextual variant scores using parameter-efficient adapters; and an openly disseminated toolkit with MIT-licensed R/Python packages, containers, DOIs, and a public API. Aim 1 develops qSuSiE for quantile-specific fine-mapping to localize variants whose effects peak in distribution tails and to test quantile-specific dominance and epistasis; Aim 2 builds qColoc and a quantile TWAS framework to integrate single-nucleus RNA-seq, chromatin accessibility, and proteomic data, including cross-cell-type interac- tion QTLs; Aim 3 adapts the Evo2-AD sequence model (parameter-efficient fine-tuning) for genome-wide variant effect scoring; Aim 4 disseminates all software, summary statistics, and an interactive web portal to maximize transparency and reuse. The project will deliver an openly browsable nonlinear AD genetic atlas, quantile-resolved maps that con- nect tail-enriched variants to brain-cell pathways, and a scalable variant scorer. Together, these resources are designed to enhance genetic risk stratification, inform clinical trial enrollment, and advance target discovery in Alzheimer's disease, facilitating community uptake via compatibility with standard fine-mapping, colocalization, and TWAS outputs.
