Fig. 6: Learned representations allow for morphological profiling of nucleoli under different perturbations.
From: Interpretable representation learning for 3D multi-piece intracellular structures using point clouds
a, q-value statistics per drug (‘Drug dataset’ of Methods) and per model indicating the confidence of each model distinguishing a given drug from control. Bar plots show 1/q for each model and drug. The y-limit is set to 100 to highlight the range of values around the 0.05 confidence threshold (dashed line at 1/q = 20). b, Three representative examples of nucleoli (GC) for the control (DMSO) and each of the 16 drugs used in this study. c,d, LDA analysis for the rotation-invariant point cloud representations of nucleoli for actinomycin D (left; n = 210) and a baseline using two random subsets of plates from the DMSO control (right; n = 140; Supplementary Note 5.5).
