(Bulik-Sullivan et al., 2015b; Lee et al., 2012). To date most of these types of analyses have been performed using genetic restricted maximum likelihood analysis (GREML) as implemented in software packages such as GCTA and LDAK (Lee et al., 2011; Speed et al., 2012; Yang et al., 2010, 2011). However, these methods require individual-level genotype data, which is often not available as most of the largest GWAS analyses are conducted through meta-analyses, and so typically only report summary results statistics (Zheng et al., 2013). Additionally GREML can be computationally prohibitive when analyzing raw genome-wide SNP data from hundreds of thousands of individuals. Consequently, most GREML analyses reported in the literature to date have been hypothesis driven studies that have involved only a small number of related traits (Table 1). Table 1.Comparison between GREML and LD Score Regression via LD HubGREMLLD Score regression via LD HubRequires individual-level dataRequires GWAS summary-level dataOne dataset at a timeIntegrates multiple GWAS results datasetsRun time depends on number of individuals and traitsRun time depends on number of traits onlyManual implementationAutomatedUsually one or a few traits at a timeMany traits simultaneouslyTypically hypothesis driven Computationally prohibitive for large numbers of individualsHypothesis driven or hypothesis-free Handles large numbers of
