Machine learning genetic screening platform for Parkin localization: proof of principle screen. (a) Screen illustration. The AI-PS platform is composed of three components i: Transduction. GFP-Parkin, pa-mCh, and dCas9-KRAB cells transduced with a subpooled library of sgRNA. ii: Machine learning modeling. Single-cell images are labeled and trained using SVM classification of mitochondrial Parkin vs. cytosolic Parkin. iii: AI-PS deployment. First, cells are imaged and segmented, and then the phenotypes of targeted cells are called and photoactivated. (b) Representative images of GFP-Parkin U2OS cells treated with DMSO or CCCP (2 h). Scale bar, 5 µm. (c) Translocation dynamic analysis of GFP-Parkin cells imaged for 25 h (1,500 min) in 1-min time lapses. Percent of cytosolic Parkin over time after 10 µM CCCP treatment supplemented with 100 nM bafilomycin A. n = 7, mean ± SD. (d) PCA analysis of 18 feature predictors calculated from 5,435 single-cell images using the R function computeFeatures from the EBImage library. The images are pooled from five different biologic repeats. (e) Table includes all the input features computed. **features selected for the model. (f) Image examples of field of view of GFP-Parkin U2OS cell screening procedure. i: Images were captured and saved on a local computer. Top: Parkin-GFP. Bottom: Draq5 dye for nucleus segmentation. ii: Cell borders were identified (green circle surrounding cell border, red circle the nucleus) following nucleus segmentation. Bottom: mCherry channel for ph-mCh protein. iii: SVM classification model was deployed and masked (red circle). Photoactivation of the SMV identified cell. Scale bar, 10 µm.