We first determined the effect that depth thresholding had on 14 649 positions probed using an Affymetrix SNP 6.0 array from genome data from the lobular breast cancer by calculating ROC curves and corresponding AUC values (Section 2) from the output of SNVMix1 at different cutoff values. The most accurate results were obtained when no depth thresholding was applied. At a threshold of 0, the AUC was 0.988 (the highest) and at a threshold of 10 reads, the AUC was 0.614 (the lowest). At a FPR of 0.01, the TPR decreased with increasing number of reads required for the threshold without exception, suggesting that depth-thresholding under the SNVMix1 model reduces overall sensitivity without increasing specificity, and should therefore be avoided. AUC for thresholds of 1, 3, 5 and 7 reads were 0.971, 0.893, 0.782 and 0.707, respectively.
