Nonlinear kernel-based high-dimensional inference for set-based genetic association studies

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Complex diseases involve nonlinear genetic effects like epistasis, which current linear models often miss.
• Existing SNP-set association tests lack power and stability for nonlinear or heterogeneous genetic data.

Purpose of the Study:

• To develop a robust, scalable framework for nonlinear, high-dimensional genetic association analysis.
• To improve the power and reliability of detecting complex genetic contributions to disease risk.

Main Methods:

• Proposed a nonlinear high-dimensional inference framework integrating scalable kernel methods.
• Utilized distance correlation-based screening, kernel PCA with Nyström approximation, and de-sparsified LASSO.
• Implemented a two-stage omnibus testing strategy for adaptive evidence aggregation.

Main Results:

• Simulations show the method maintains Type I error control and higher power than existing tests, especially for nonlinear effects.
• The framework outperforms Sequence Kernel Association Test and adaptive Sum of Powered Score tests in nonlinear scenarios.
• Identified gene associations with Alzheimer's disease brain volumes linked to neuronal excitability and calcium signaling.

Conclusions:

• The developed framework offers a powerful and scalable solution for nonlinear set-based inference in genome-wide studies.
• This expands the analytical tools for understanding complex genetic architectures in diseases.
• The approach is particularly valuable for dissecting genetic contributions to neurodegenerative disorders.