Finally, we used summary statistics to create genome-wide polygenic scores (PGS) in the S4S sample. As our discovery sample included individuals of EA and AA ancestry, we selected the same ancestral populations in S4S for our validation effort (62% Female; European ancestry N = 2,761; African ancestry N = 1,175). We conducted all analyses separately by ancestry using ancestry specific GWAS results for constructing PGS. In order to maximize available sample size, we ran 10 additional GWAS in S4S using a leave-one-out (LOO) strategy in which 10% of each sample was omitted from the GWAS. PGS were then constructed from the meta-analyzed results of the ancestry specific GWAS in COGA and the S4S sample in which each hold out was not included, similar to previous GWAS (Otowa et al., 2016). For example, the PGS for the first 10% of EA removed in S4S were constructed from GWAS weights of the EA results in COGA meta analyzed with GWAS weights of the remaining 90% of S4S respondents using a sample-size based meta analysis (Willer et al., 2010)
