Fig. 2
From: ACE: a probabilistic model for characterizing gene-level essentiality in CRISPR screens
Detection of Absolute Essentiality in Simulated Data. A Inferred essentiality values (\(\widehat \phi \)) for 300 genes simulated in three replicates at each essentiality value shown (ϕ). Simulated data sets also included 3150 nonessential genes (ϕG=0; not shown), 300 of which were provided to each method as negative controls. Black horizontal bars indicate true values (\(\widehat \phi = \phi \)). B Performance in binary classification (essential vs. nonessential) of simulations of 300 “essential” genes at various values of ϕG and 300 “nonessential” genes. Performance reported as the Area under the Receiver Operating Characteristic (ROC) curve. Only initial and final read counts from the simulation were used for all methods. C Full ROC curves for each of the simulated essentiality values. “ACE” — our probabilistic method (thresholded on log likelihood ratios), “AFC” — method based on average fold changes in sgRNA abundance (thresholded on z-scores; see the “Methods” section). “BAGEL” — Bayesian Analysis of Gene Essentiality [2] (thresholded on reported Bayes Factors); “JACKS” — Joint analysis of CRISPR/Cas9 knockout Screens [8] (thresholded on reported p-values). “CRISPhieRmix” — hierarchical mixture model [23] (thresholded on reported essential gene FDR)