Figure 4. Receiver Operating Characteristic (ROC) curve for 10 fold cross-validation.

A) Discovery data (32,587 SNPs) B) Combined data (32,375 SNPs). Ten sets of training and testing subsets were created. Multi-SNP models were developed by Bayes C analysis in training sets and were validated in testing set. Each broken line represents one model and solid bold line is the average area under curve (AUC) of all models. AUC is equivalent to the probability of correctly assigning a random pair of observations (positive and negative) to case and control. The diagonal represents a model with no predictive ability (AUC  = 0.5).