While comparing with the SNP array data, we defined a true positive (TP) SNV as an ab or bb CRLMM genotype. A true negative (TN) SNV was defined as an aa genotype from the SNP array. For the Sanger validated positions, a TP was an SNV that was confirmed by Sanger sequencing, whereas a TN was a position that was not confirmed. To evaluate our models against these data, we computed p(SNVi)=γi(ab)+γi(bb) and standard receiver operator characteristic (ROC) curves. The area under the ROC curve (AUC) was computed as a single numeric metric of accuracy that effectively measures the trade-off between sensitivity and specificity. As an additional measure, we computed the F-statistic: , where precision is measured as the proportion of predictions that were true and recall is the proportion of true SNVs that were predicted.
