A Novel, Variance Component-Based Method for Detecting Brain-Behavior Associations in Neuroimaging Data.

Area of Science:

• Genetics
• Biostatistics
• Neuroimaging

Background:

• The Sequence Kernel Association Test (SKAT) is a standard method for high-dimensional genetic association studies.
• SKAT's omnibus nature can limit the interpretability of results, particularly in identifying specific signal patterns.
• Existing methods may struggle to differentiate between global and local signals effectively.

Purpose of the Study:

• To develop a novel statistical method, LaxKAT (linear maximum kernel association test), for enhanced signal detection in high-dimensional data.
• To improve upon the interpretability and power of existing association tests like SKAT.
• To identify sex-specific differences in brain cortical thickness patterns using neuroimaging data.

Main Methods:

• Developed LaxKAT, which maximizes the SKAT statistic over a defined subspace of linear kernels.
• Conducted simulation studies to evaluate LaxKAT's performance against existing methods.
• Applied LaxKAT to neuroimaging data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.

Main Results:

• LaxKAT demonstrated improved global and local power compared to previous methods in simulations.
• The method successfully controlled the family-wise error rate (FWER).
• Analysis of ADNI data identified specific brain regions with sex-specific cortical thickness variations.

Conclusions:

• LaxKAT offers a powerful and interpretable alternative for analyzing high-dimensional genetic and neuroimaging data.
• The method enhances the ability to detect both broad and localized signals.
• LaxKAT provides a valuable tool for identifying complex biological patterns, such as sex differences in brain structure.