Multiple comparison procedures for neuroimaging genomewide association studies

Abstract

Recent research in neuroimaging has focused on assessing associations between genetic variants that are measured on a genomewide scale and brain imaging phenotypes. A large number of works in the area apply massively univariate analyses on a genomewide basis to find single nucleotide polymorphisms that influence brain structure. In this paper, we propose using various dimensionality reduction methods on both brain structural MRI scans and genomic data, motivated by the Alzheimer's Disease Neuroimaging Initiative (ADNI) study. We also consider a new multiple testing adjustment method and compare it with two existing false discovery rate (FDR) adjustment methods. The simulation results suggest an increase in power for the proposed method. The real-data analysis suggests that the proposed procedure is able to find associations between genetic variants and brain volume differences that offer potentially new biological insights.

Equivalence of kernel machine regression and kernel distance covariance for multidimensional phenotype association studies

“…Finally, several works have utilized the KDC/KMR family members in applications that include genetic pathway analysis using KMR Liu et al (), voxel‐wise genome‐wide association studies using least squares KMR (Ge et al, ), neuroimaging genome‐wide association using DC (Hua et al, ), and multiple change point analysis using DC (Matteson and James, ). Some recent studies have presented and discussed the equivalence between these statistics, such as distance‐based permutation test for between group comparisons from Reiss et al (), the relationships between Genomic Distance‐Based Regression and KMR from Pan () and the equivalence between DC and HSIC from Sejdinovic et al ().…”

“…The average voxel value of each region was used to represent each of the 119 ROIs. With these 119 regions, Hua, Nichols, and Ghosh () used the DC test and discovered that the difference in brain volumes were highly associated with a common variant in the intron region of gene FLJ16124 , with a total of 141 SNPs within gene FLJ16124 that were identified by the SNP‐gene mapping from Hibar et al (). The subject pool consists of 741 subjects from the ADNI study that have passed the quality control filtering according to Stein et al (), which we retained for the simulation and real data analysis.…”

Generalized reduced rank latent factor regression for high dimensional tensor fields, and neuroimaging-genetic applications

“…For example, the probabilistic formulations do not scale well with dimensionality; and standard brute force massive univariate approaches (Stein et al, 2010a; Vounou et al, 2012) treat each voxel and predictor as independent units and compute pairwise significance, and the loss of spatial information and the colossal multiple comparison corrections involved have high costs in terms of sensitivity (Hua et al, 2015). Various attempts have been made to remedy this.…”

Random Projection for Fast and Efficient Multivariate Correlation Analysis of High-Dimensional Data: A New Approach

“…Using the same data set of genome-wide SNPs and whole-brain neuroimaging voxels from the ADNI, in a recent study, Hua et al ( 2015 ) performed dimensionality reduction by selecting 119 brain regions of interest based on an anatomical brain atlas. Distance covariance (Székely et al, 2007 ) was applied to infer the relationship between the single SNP predictors from the entire genome and the average voxel values at the 119 brain regions, utilized as multivariate response.…”

“…Distance covariance (Székely et al, 2007 ) was applied to infer the relationship between the single SNP predictors from the entire genome and the average voxel values at the 119 brain regions, utilized as multivariate response. In order to overcome the multiple testing problem, Hua et al ( 2015 ) also introduced a local false discovery rate (FDR) modeling algorithm. The authors showed that using their method, they were able to find 23,128 significant SNPs at α-level 0.05, while simple linear regression yielded no significant SNPs.…”