Fig. 2.

Performance of Lancet and other methods on the virtual tumors. a Precision/recall curves for somatic SNVs called by Lancet, MuTect, MuTect2, LoFreq, Strelka, and Strelka2 on the virtual tumor. Curves are generated by sorting the variants based on the confidence or quality score (QUAL) assigned by each tool. Each point on the curve corresponds to precision and recall for all the SNVs with confidence score greater than or equal to a specific quality threshold. The curve for an ideal tool (with no errors) should start from the top left corner (with precision = 1) and produce a straight horizontal line. Any deviation from a straight line is due to errors introduced by the variant calling process. Specifically, deviations at low recall rates are indicative of low performance of the scoring system adopted by the tool (false positive variants reported with high score). b Precision/recall curves for somatic indels called by Lancet, MuTect2, LoFreq, Strelka, and Strelka2 on the virtual tumor. Number of true-positive (c) SNVs and (d) indels at different variant allele fractions for each method compared to the truth call set