Fig. 2

Evaluation of FIREVAT variant refinement performance on real-world datasets against ground truth. a FIREVAT variant refinement performance on 360 MC3 samples with known ground truth data against three other manual hard-filtering approaches: Lancet (light blue: LAN-F), MuTect (green: MUT-F), and Varscan (navy: VAR-F). FIREVAT refinement results yielded the highest F1 score when evaluated on combined callsets. b Scatterplot of the specificities and the initial sum of artifact signature weights for FIREVAT and the other filtering approaches. FIREVAT refinement specificity showed a positive correlation with the initial sum of artifact signature weights. c Variant refinement was performed using FIREVAT and DToxoG on the WES data of 6 breast cancer samples with technical replicates (n = 12). We used five different objective functions that assign different weights to each of the four terms constituting the objective value. When evaluated against the ground truth data, FIREVAT achieved the highest precision level and F1 score against DToxoG