Fig. 3. MFN2 Genotype Prediction does not depend on specific features.

a Histograms of AUCs resulting from models built from combinations of a set of 8 features used when picking, arranged in panels split by the number of features used in each subset. b Bar chart of the Kruskall–Wallis χ² values for the 16 features with the highest values. “Mito” and “Nuc” indicate features measured from mitochondrial and nuclear stains, respectively. c Bar chart of the AUCs resulting from models built out of each individual feature. Note that the most significant feature was unable to produce a good model alone. d Bar charts of the AUCs resulting from models built using all but one feature (leaving 15 features). All AUCs listed indicate the performance of the model trained from labeled data on their ability to predict unlabeled cells in admixed conditions.