Extended Data Fig. 5: Comparison of CSL-derived nuclear speckles/PML bodies with supervised segmentation of nuclear speckles/PML bodies.

a: CSLs generated using CAMPA for an example 184A1 nucleus. b: Measured intensities of canonical nuclear speckle markers SON and SRRM2. c: Probability maps of nuclear speckles generated through supervised pixel clustering of single-channel images in Ilastik. d: Nuclear speckle classifier derived from thresholding the probability map at P=0.95. e: Comparison of pixels assigned to nuclear speckles using supervised classifiers based on SON, SRRM2 with nuclear speckle CSL. f: Quantitative comparison of the different classification approaches using SON and SRRM2 classifiers as the ground truth. x-axis shows the value at which the probability map is thresholded to generate the segmentation. Dashed line at P=0.95 is the classification shown in d. Metrics computed on a random 10% subsample of all data. g: F₁-scores for each comparison for all perturbations individually. Data from all pixels in a single well for each condition. Probability map threshold at P=0.95. h-j: As in b-d for canonical PML body markers SP100 and PML. Classification in i is the probability map thresholded at P=0.6. k: As in e for PML bodies. l: As in f for PML bodies. Dashed line at P=0.6 is the classification shown in j. m: As in g for PML bodies. Probability maps thresholded at P=0.6.