We generated true genetic liability for all European, African, and admixed individuals within each simulation trial.2 Briefly, m variants evenly spaced throughout the simulated genotypes were selected to be causal, and the effect sizes (β) were drawn from a normal distribution β∼N(0,(h2/m)), where h2 is the heritability.2 Constant heritability and complete sharing of effect sizes in African ancestry and European ancestry individuals was assumed. The true genetic liability was computed as the summation of all variant effects multiplied by their genotype for each individual (X=∑i=1mβmgm) standardized to ensure total variance of h2 (G=X−μXσX∗h2). Finally, the non-genetic effect (ε=N(0,1−h2)) standardized to explain the remainder of the phenotypic variation (E=(ε−με/σε)∗1−h2) was added to the genetic risk, defining the total trait liability (G+E).2 Cases were selected from the extreme tail of the liability distribution, assuming a 5% disease prevalence. An equal number of controls and 5,000 testing samples were randomly selected from the remainder of the distribution; all 5,000 admixed individuals were also used for testing. Across simulation replicates we varied causal variants (m = [200, 500, 1,000]) and trait heritability (h2 = [0.33, 0.50, 0.67]); however, for simplicity, main text results assume m = 1,000 and h2 = 0.50.
