TFEB screen statistical power modeling and performance. (a) GFP-TFEB phenotype classification performance by CNN. Single-cell images of GFP-TFEB induced with the designated library (color coded, H1–H7) were collected. Precision-recall curve from 5,203 single-cell images pooled from three biologic repeats. Image collection began 8 h after starvation was initiated and continued for another 10 h. The accuracy was computed from the integral area under the precision-recall curve (area under the curve [AUC]). The AUC was calculated per subpooled library (designated by color). (b) i-vii: Statistical power analysis by simulation of the sgRNA read counts retrieved from the TFEB screen. Mean unique sgRNAs detected used for the simulation was 2,680 ± 512 and average read count of 282 ± 64. The simulation was done using the R package PROPER, effect size (log of fold-change) is shown in the x-axis, power in the y-axis, the curved lines are color coded for number of biologic repeats. For each subpooled library, there was a sample size of 3 (red curve), 5 (green curve), 7 (cyan curve), and 9 (purple curve).