A Spatial-Correlated Multitask Linear Mixed-Effects Model for Imaging Genetics.

Area of Science:

• Neuroimaging
• Genetics
• Biostatistics

Background:

• Imaging genetics links brain imaging quantitative traits (QTs) with genetic markers (SNPs) to understand complex diseases.
• Existing linear models often overlook the spatial interdependencies among QTs, potentially limiting the efficiency of information sharing across brain regions.

Purpose of the Study:

• To develop a novel Bayesian regression framework for imaging genetics that explicitly models spatial dependencies between QTs.
• To identify significant associations between QTs and genetic markers while leveraging the inherent structure of brain imaging data.

Main Methods:

• Developed a spatial-correlated multitask linear mixed-effects model to capture dependencies between QTs.
• Implemented the model within a Bayesian framework using a Markov chain Monte Carlo (MCMC) algorithm for inference.
• Utilized MCMC samples with the Cauchy combination test to assess SNP-QT associations, overcoming multitest challenges.

Main Results:

• Simulation studies demonstrated superior statistical power compared to classical models that do not account for QT interdependencies.
• The proposed model effectively identified significant associations between SNPs and QTs by incorporating spatial correlations.

Conclusions:

• The novel spatial Bayesian regression framework enhances the detection of genetic associations with brain imaging QTs.
• This approach offers valuable insights into the pathogenesis of complex diseases like Alzheimer's disease by modeling inter-QT relationships.