Fig. 3: Biological applications with ECLiPSE in 3D SMLM.
a, Representative images for 3D data color coded according to depth using the color scale bar (lysosomes (left), healthy mitochondria (middle) and depolarized mitochondria (right)). The insets show zoomed-in 3D views of the regions within the white boxes. b, Data exploration using PCA after feature extraction using ECLiPSE shows a clear separation between lysosomes and mitochondria (left), but a nonnegligent amount of similarities between healthy and depolarized mitochondria (right). c, The classification results using the partial least squares classifier on the variable selected lysosome versus mitochondria data (left; 98.6 ± 0.1% accuracy) and the random forest classifier on the variable selected healthy versus depolarized mitochondria (right; 75.8 ± 0.6% accuracy). d, The quantification of four biological properties of healthy and depolarized mitochondria indicating that two nonvariable selected properties do not show significant differences (number of localizations and boundary surface curvature (two left-most graphs)) and two variable selected properties show significant differences (major axis and sphericity (two right-most graphs)). For classification in c, only the results obtained by the 100 best models out of 1,000 generated models is shown. Lysosomes: n = 11 cells (three biological replicates); healthy mitochondria: n = 16 cells (four biological replicates); depolarized mitochondria: n = 9 cells (three biological replicates). In d the black line represents the median of the shown biological property and 1% upper and lower values were removed only for visualization purposes. P values were calculated using a two-sided Wilcoxon rank sum test.
