Fig. 4

Precision vs Recall for NA12878 using benchmarks from exome sequencing GIAB Indels (a, c) and SNPs (b,d). Plots a) and b) show the full precision vs recall for each method. Solid lines represent EAGLE’s likelihood. Dotted lines represent the caller’s quality score. Recall levels are shown in increments of 50 variant calls with the maximum level based on the number of variants in the GIAB benchmark set. The variant calls were ranked based on our model’s marginal posterior probability or each caller’s quality score respectively. Precision is the fraction of high ranking variants which are correct, plotted over a range of thresholds. Plots c) and d) show the best precision at a given recall among all methods with EAGLE versus among all methods without EAGLE, for indels and SNPs respectively