Fig. 4

gscreend increases phenotype detection accuracy. agscreend analysis workflow. b–e Comparison of gene ranking by BAGEL, CRISPhieRmix, CRISPRBetaBinomial, MAGeCK, ScreenBEAM, and gscreend for CRISPR knockout screen performed in HCT116 cells [2]. b Recall at 95% precision (analysis as in Fig. 1d) for 1 to 3 biological replicates of timepoint T18. Essential and non-essential genes were defined according to Hart et al. [25, 42]. c Ranking of gene encoding ribosome components (structural constituent of ribosome - GO:0003735) by the six different analysis tools and using all 3 biological replicates. d Volcano plots illustrating gscreend, mageck, and BAGEL analysis results for gene encoding ribosomal components (red) and non-essential genes [25] (blue). Horizontal lines indicate FDR thresholds of 1%. e Log2 normalized abundances of gRNAs targeting the 5 selected genes at time point T0 and 3 replicates of time point T18. f–g Recall at 99% precision by CRISPhieRmix, CRISPRBetaBinomial, gscreend, MAGeCK, gscreend, and ScreenBEAM for simulated data of 1 to 3 biological replicates. Other simulation parameters: library distribution width 7.5, cell splitting coverage 200, doubling time 30 h. Precision-recall curves were calculated for detection of essential (f) and growth-suppressing (g) genes. h Recall at 99% precision of essential genes for simulations with different library width and cell splitting coverage. Genes were ranked using gscreend (left) or MAGeCK (right). i Recall at 95% precision for gscreend and MAGeCK analysis of screens performed either using a cell splitting coverage of 200 and the TKO_v1 library (library distribution width ca. 7.8) [2] or a cell splitting coverage of 500 and the Avana library (library distribution width ca. 4.6) [40]. Essential and non-essential genes were defined according to Hart et al. [25, 42]