Finally, we have used MuTect in several recent studies and found a consistent validation rate of ~95% in coding regions based on multiple orthogonal validation technologies3,4,7,19,24–30 (Table 2) These studies used earlier versions of MuTect which were less sensitive, however a recent publication13 using this version was able to detect mutations present at 7% allelic fraction (8 reads out of 102) which were subsequently validated by ultra-deep sequencing (~6,000x). In fact, the validation rate is not the best measure for comparing false positive rates across studies because it depends on the ratio of false positive to true mutations, which varies across tumor types. We therefore also report the false positive rate itself (Table 2). We observe a median false positive rate of 0.16/Mb, which is lower than the rate we report using whole genome data (Fig. 3) but consistent with the rate measured when restricting to coding regions (Supplementary Fig. 5), indicating that coding regions are less prone to sequencing and alignment errors.
