A General Framework of Brain Region Detection And Genetic Variants Selection in Imaging Genetics

TL;DR

This paper introduces a novel statistical framework combining canonical correlation analysis and scalable algorithms to jointly analyze brain imaging, genetic data, and phenotypes, effectively identifying significant brain regions and genetic variants linked to cognitive functions.

Contribution

It presents a new integrated modeling approach for imaging genetics that addresses high-dimensional data, spatial information, and computational scalability, enabling simultaneous detection of relevant brain regions and genetic variants.

Findings

Identified significant association between the caudate nucleus and reaction speed.

Discovered specific SNPs and genes linked to cognitive function.

Demonstrated scalability and effectiveness on UK Biobank data.

Abstract

Imaging genetics is a growing field that employs structural or functional neuroimaging techniques to study individuals with genetic risk variants potentially linked to specific illnesses. This area presents considerable challenges to statisticians due to the heterogeneous information and different data forms it involves. In addition, both imaging and genetic data are typically high-dimensional, creating a "big data squared" problem. Moreover, brain imaging data contains extensive spatial information. Simply vectorizing tensor images and treating voxels as independent features can lead to computational issues and disregard spatial structure. This paper presents a novel statistical method for imaging genetics modeling while addressing all these challenges. We explore a Canonical Correlation Analysis based linear model for the joint modeling of brain imaging, genetic information, and clinical phenotype, enabling the simultaneous detection of significant brain regions and selection of important genetic variants associated with the phenotype outcome. Scalable algorithms are developed to tackle the "big data squared" issue. We apply the proposed method to explore the reaction speed, an indicator of cognitive functions, and its associations with brain MRI and genetic factors using the UK Biobank database. Our study reveals a notable connection between the caudate nucleus region of brain and specific significant SNPs, along with their respective regulated genes, and the reaction speed.