We evaluated performance of the models on a lobular breast cancer sample sequenced to >40× haploid coverage (Shah et al., 2009b). In addition, we compared results obtained from the same genome at 10× coverage. We first trained the model using 14 649 protein coding positions for which we generated matching Affymetrix SNP6.0 calls. We computed the AUC for SNVMix1, SNVMix2 and Maq. Table 2 shows that the highest AUCs were obtained with SNVMix2 on the 40× genome, followed by SNVMix2 on 10× genome (AUCs of 0.9929 and 0.9905, respectively). Both of these were higher than results achieved for SNVMix1 (AUC of 0.9880) and Maq (0.9824 for 40× and 0.9115 for 10×—both for the r=0.001 parameter setting). After fitting the model to the 14 649 positions, we evaluated the performance using 497 candidate mutations originally detected using SNVMix1 at 10×, which were validated using Sanger amplicon sequencing (Section 2). These consisted of 305 true SNVs (variants seen in the Sanger traces) and 192 that could not be confirmed in the Sanger traces. Table 2 shows the sensitivity, precision and F-measure results
