Interpretable High-Dimensional Inference Via Score Projection with an Application in Neuroimaging

Area of Science:

• Neuroimaging
• Genetics
• Biostatistics

Background:

• High-dimensional data in neuroimaging and genetics often requires summary measures for association testing.
• Existing methods can lack power to localize specific regions despite significant overall associations.
• Underpowered localization tests hinder the identification of key biological markers.

Purpose of the Study:

• To develop a novel statistical method for localizing associations in high-dimensional neuroimaging and genetic data.
• To generalize Rao's score test for improved signal detection and localization.
• To reduce the number of statistical comparisons for enhanced power.

Main Methods:

• Proposed a generalization of Rao's score test by projecting the score statistic onto a linear subspace.
• Developed a method to localize signals within high-dimensional parameter spaces.
• Reduced degrees of freedom for inference, aligning with the score test's dimensionality.

Main Results:

• Simulation studies showed the proposed test has competitive power compared to existing methods.
• The method effectively localizes signals in high-dimensional spaces.
• Analysis of the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset was performed.

Conclusions:

• The generalized Rao's score test offers a powerful approach for neuroimaging and genetic association studies.
• Cortical thinning in the frontal and temporal lobes may serve as a significant biomarker for Alzheimer's disease risk.
• The method facilitates more precise identification of disease-related biological markers.