ENIGMA, to date, has used a mass-univariate analysis, where each trait (or brain measure) is considered on its own, and each genetic variant is considered on its own. Recent multivariate analyses can cluster voxels in the brain—or SNPs on the genome—to empower analyses, sometimes with both forms of clustering occurring at once (Hibar et al. 2011a; Thompson et al. 2013). Some of these multivariate analysis methods have been used to detect significant hits in image-wide genome-wide searches in cohorts of under 1,000 subjects (Ge et al. 2012; Chen et al. 2012b; Jahanshad et al. 2013b). In most analyses, multivariate refers to condensing information on the imaging side, not the genetic side, although both methods and joint methods are emerging. Multivariate methods can be quite sophisticated mathematically. Some draw upon a century of powerful work in classical quantitative genetics and twin designs. Chiang and colleagues (Chiang et al. 2011, 2012), for example, computed the “cross-trait cross-twin correlation” between all pairs of voxels in an image, to pull out “image clusters” with common genetic determination (see also Chen et al. 2011, 2012a).
