Each of the five clinical phenotypes was regressed on each of the 527,828 autosomal SNPs, using the demographic and principal component scores described above as covariates. These regressions were fit using a Rapid Feasible Generalized Least Squares (RFGLS) algorithm, which we developed to efficiently account for the unique familial clustering of the MCTFR data (Li, Basu, Miller, Iacono, & McGue, 2011). Briefly, the RFGLS algorithm takes into account both genetic and environmental contributions to phenotypic similarity among family members by modeling the covariance structure separately in the twin, sibling, and adoptive families. In the regression analysis, the SNP effect was modeled in terms of both an additive effect of the number of minor alleles (i.e., the main effect) and interaction between number of minor alleles and generation (i.e., to determine whether SNP effects differed in the parent and offspring generations.)
