Figure 1.

Receiver operating characteristic area under the curves ROC AUCs and predictive r² for various predictive methods in three case studies training polygenic predictive models using meta‐GWAS summaries. For type 1 diabetes, the T1DGC (n = 8,005) was used for training and the Welcome Trust Case Control Consortium for validation. For coronary artery disease (CAD), cardiogram (n = 137,535) was used for training and the WTCCC for testing. For schizophrenia, the Psychiatric Genomics Consortium (n = 74,511) was used for training and the MGS study for testing. For each analysis and method, results are presented for the best performing sparsity. ROC AUC 95% confidence intervals were calculated using 2,000 stratified bootstrap replicates