SVM classification build for Parkin screen. (a) Two-dimensional representation of nonlinear hyperplane separation of mitochondrial (mito) Parkin phenotype vs. cytosolic (cyto) phenotype. Mitochondrial Parkin predicted cells are in red, cytosolic Parkin in black. The variable o represents correct classification, and x misclassification. (b) The confusion matrix is computed from a test set composed of three biological replicates pooled single-cell images (n = 4,894). Prior to the SVM classification prediction, the images were manually labeled. The green boxes are designated for the true positive single-cell counts (upper left) and the true negative (lower right). The pink boxes represent the false positive single-cell counts (upper right) and the false negative (lower left). (c) Precision recall plot (right) summarizes the same image test set. The accuracy is calculated from the precision recall curve. (d) Beeswarm plot summarizing single-cell TFEB-GFP segmentation score comparison. The image contains 79 cells from the TFEB screen, segmented with AI-PS segmentation procedure (using the R package EBImage; black) or Cellprofiler segmentation pipeline (red). Mean ± SD. (e) dCas9-KRAB–expressing U2OS cells treated with sgRNA targeting either TRANS or CDH2 and immunostained using TRANS or CDH2 antibodies. Scale bar, 20 µm. AUC, area under the curve; IOU, intersection over union.