Fig. 1

Logistic regression model with ten-fold validation. By harnessing L1 penalty (the lasso), we further removed irrelevant SNPs in fold2 after the preselection step in fold1. Smaller lambda (the penalty parameter) values correspond to fewer SNPs removed, and numbers on the top of the plot indicate how many SNPs survived with respect to specific lambdas as X-axis (natural logarithm scale). We estimated the mean and standard error (SE) for AUCs across 100 different lambda values, and reported the largest lambda such that AUC is within 1 SE of the optimum (the left vertical dashed line shows the lambda with maximum of AUC, while the right vertical dashed line shows the lambda with AUC being within 1 SE of that maximum). The optimal 10-fold cross-validated AUCs on fold 2 data was 0.673 with regularization parameter lambda of 0.00954