Large-scale distributed analyses of over 30,000 magnetic resonance imaging scans recently detected common genetic variants associated with the volumes of subcortical brain structures. Scaling up these efforts, still greater computational challenges arise in screening the genome for statistical associations at each voxel in the brain, localizing effects using "image-wide genome-wide" testing (voxelwise genome-wide association studies, vGWASs). Here we benefit from distributed computations at multiple sites to metaanalyze genome-wide image-wide data, allowing private genomic data to stay at the site where it was collected. Site-specific tensor-based morphometry is performed with a custom template for each site, using a multichannel registration. A single vGWAS testing 107 variants against 2million voxels can yield hundreds of terabytes (TB) of summary statistics, which would need to be transferred and pooled for metaanalysis. We propose a two-step method, which reduces data transfer for each site to a subset of single-nucleotide polymorphisms and voxels guaranteed to contain all significant hits.

Big data, GWAS, Metaanalysis, Multiple comparisons correction, Multisite, Neuroimaging genetics,
no subscription
Erasmus MC: University Medical Center Rotterdam

Jahanshad, N, Roshchupkin, G.V, Faskowitz, J. (Joshua), Hibar, D.P. (Derrek P.), Gutman, B.A. (Boris A.), Adams, H.H.H, … Thompson, P.M. (Paul M.). (2017). Multisite Metaanalysis of Image-Wide Genome-Wide Associations With Morphometry. In Imaging Genetics (pp. 1–23). doi:10.1016/B978-0-12-813968-4.00001-8