The standardized residuals were then used as the final simulated trait. Samples of size 50,000 and 200,000 individuals were randomly selected as the base sample and used to generate the GWAS summary statistics. Then 100, 1,000, 10,000, and 100,000 samples independent from the base were randomly selected as the target sample. PRS analyses were then performed on these base and target data using the latest version of lassosum (v0.4.4), LDpred (v1.0.6), and PRSice 2 (v2.2.1), on servers equipped with 286 Intel 8168 24 core processors at 2.7 GHz and 192 GB of RAM. Default parameters of each program were used. The run time and memory usage of each program were measured using the Linux "time" command and the predictive power of the methods was assessed according to phenotypic variance explained (R2). The entire process was repeated 10 times to obtain an estimated distribution of run time, memory usage, and predictive power.
