Next, to apply the down-sampling benchmark, we used 3,753 validated somatic mutations, stratified by allele fraction (median=0.28, range=0.07–0.94), in colorectal cancer7 with deep-coverage (≥100x) exome-capture sequencing downloaded from dbGAP (phs000178). Finally, to apply the virtual tumor benchmark, we used deep-coverage data from two high coverage whole-genome samples (NA12878 and NA12981) sequenced on Illumina HiSeq instruments by the 1000 Genomes Project42 and another previous study43, across 1 Gb of genomic territory. Note that we cannot use the Panel of Normals filter (HC+PON) in the virtual tumor sensitivity benchmark, because it discards common germline sites.
