Fig. 4

Classification of loss-of-function variants using efficiency-corrected scores. a–d Histograms of score distributions for variants tested in four screens: BRCA1-AT-NGG (a), BRCA1-CG-NGG (b), BRCA2-AT-NGG (c), BRCA2-CG-NGG (d). Blue bars represent variants that are classified as loss-of-function variants using thresholds determined by Gaussian mixture modeling. e Ratio of loss-of-function variants in different mutational categories. f Ratio of loss-of-function variants within different ClinVar interpretation categories. g ROC curves for loss-of-function variant classification using our efficiency correction model and MAGeCK. The orange and light blue lines used 49 “pathogenic/likely pathogenic” variants in ClinVar as true positives and 190 “benign/likely benign” variants as true negatives. The red and blue lines used 37 “pathogenic/likely pathogenic” variants and 146 “benign/likely benign” variants, after filtering the sgRNAs that had editing activity lower than 0.2. h Fold change of sgRNAs for 7 loss-of-function variants in the screens, as well as the fold change of three control variants in Fig. 1b, is plotted against the fold change of cell growth in the cell growth assays. Pearson’s correlation coefficient is shown. Detailed data for the figure is presented in Table S1