In the framework considered here, a set of genetic markers is genotyped on an initial training sample and each marker is tested for association to a trait. Effect sizes are estimated for each marker and used to construct a polygenic score for each subject in an independent replication sample. The score is tested for association in the replication sample, in which the tested trait may differ from that in the training sample. The correlation and mean square error between the polygenic score and the tested trait are calculated. If the traits are binary, the AUC is obtained.
