Co-sparse reduced-rank regression for association analysis between imaging phenotypes and genetic variants

Area of Science:

• Neuroimaging Genetics
• Statistical Genetics
• Computational Biology

Background:

• Understanding inherited neuropsychiatric disorders requires analyzing genetic variants and brain imaging phenotypes.
• Traditional methods face computational challenges and overlook complex correlations due to high-dimensional data.
• Existing multivariate methods often use l1 penalties, leading to potential over-selection of variables.

Purpose of the Study:

• To develop a novel statistical model addressing computational and statistical challenges in imaging genetics.
• To identify complex, many-to-many correlations between genetic variants and imaging phenotypes.
• To improve the accuracy and efficiency of variable selection in neuroimaging genetic studies.

Main Methods:

• Proposed a co-sparse reduced-rank regression model for dimensional reduction.
• Developed an iterative algorithm using a group primal dual-active set formulation.
• Employed a non-convex penalty for simultaneous detection of important genetic variants and imaging phenotypes.

Main Results:

• Simulation studies demonstrated accurate and stable performance in parameter estimation and variable selection.
• The novel method efficiently and precisely identified key genetic variants and imaging phenotypes.
• Successfully detected novel Alzheimer's disease-related genetic variants and regions of interest in a real-data application.

Conclusions:

• The proposed co-sparse reduced-rank regression model offers a valuable statistical toolbox for imaging genetic studies.
• The method effectively handles high-dimensional data and complex correlations.
• It provides a more precise approach to variable selection compared to existing methods.