Hyper-graph based Adaptive Sparse Multi-view Canonical Correlation Analysis with Application to Neuroimaging Genetics Study of Alzheimer’s Disease

Abstract

Neuroimaging genetics has gained more and more attention on account of detecting the linkage between the brain imaging phenotypes (i.e., regional volumetric measures) and the genetic variants (i.e., Single Nucleotide Polymorphism (SNP) in Alzheimer’s disease (AD)). To overcome the problem of sparse multi-view canonical correlation (SMCCA) ‘unfair combination of pairwise convariance’, introducing adaptive weights when combining pairwise covariances, a novel formulation of SMCCA, named adaptive SMCCA. In this paper, we integrate multi-modal genomic data from postmortem AD brain and proposed a hyper-graph based sparse multi-view canonical correlation analysis (HGSMCCA) method to extract the most correlated multi-modal biomarkers. Specifically, we utilized the adaptive sparse multi-view canonical correlation analysis (AdsSMCCA) framework, consider the benefit of hyper-graph-based regularization term into consideration that will contribute to the selection of more discriminative biomarkers. We propose a hyper-graph optimization strategy based on the adaptive SMCCA model, and we apply it to neuroimaging genetics data. All these results demonstrate the capability of HGSMCCA in identifying diagnostically genotype-phenotype patterns.