to performance in the matched European ancestry sample from the same study, by dividing effect sizes (i.e., non-European ancestry/European ancestry). Odds ratios were first converted to log(OR). We multiplied values by 100 so that performance for each non-European ancestry sample could be expressed as a percentage of European ancestry performance, which was standardized to 100%. For example, the first polygenic scoring study of schizophrenia12 found that polygenic scores explained only 0.4% of phenotypic variance in an African ancestry sample, whereas 3.2% of phenotypic variance was explained in a matched European ancestry sample. Consequently, the value of 12.5% ((0.004/.032) × 100) is represented as the top-most yellow point in Fig. 2 (see schizophrenia and Purcell 2009 in the citation information for this point). By normalizing within-study effect sizes to European ancestry effect sizes, we were able to combine observations across phenotypes, and therefore to obtain general estimates of polygenic score performance across ancestry groups and complex genetic phenotypes.
