Our data suggest that MuTect has an advantage over other methods in terms of its tradeoff between specificity and sensitivity (Fig. 4). The advantage in sensitivity of MuTect is derived from the variant detection statistical test, which includes an estimation of the allele fraction of the event, and the working point chosen along the ROC curve. SomaticSniper and JointSNVMix use a model based on a clonal mutation in a pure, diploid tumor (and thus assume a fixed 50% allele fraction). This assumption reduces sensitivity for lower allele fraction events. In contrast, Strelka specifically considers allele fraction, and thus in the STD configuration has similar sensitivity to MuTect. However, when running in the recommended HC configurations to control false positives, MuTect has only a minor drop in sensitivity compared with the other methods. This is likely because the filters in MuTect were carefully tuned to reject true false positive calls without sacrificing sensitivity.
