Fig. 2.

F1 score for the benchmark of the DelSIEVE model. Varying are the mutation rate (the horizontal axis), the relative deletion rate (the vertical strip), the coverage quality (the horizontal strip), and the simulated dropout type (the shaded or blank boxes). Each simulation is repeated $n=10$ times, with each repetition denoted by colored dots. The gray dashed lines represent the optimal values of each metric. Box plots comprise medians, boxes covering the interquartile range (IQR), and whiskers extending to 1.5 times the IQR below and above the box. Data points were removed if the proportion of simulated ground truth was less than 0.1%. Both DelSIEVE and SIEVE were configured to match the dropout mode (ADO or LDO) employed during the simulation process. Box plots of the F1 score for calling alternative-remaining single deletion (a), reference-remaining single deletion (b), double deletions (c), single mutants (d), and double mutants (e). The results in c when mutation rate was $8\times {10}^{-6}$ were omitted as very few double deletions were generated (less than 0.2%; see Simulation design section in Additional file 1: Supplementary notes). f Box plots of the F1 score for calling all types of mutations considered in a–e