PB-scope: Contrastive learning of dynamic processing body formation reveals undefined mechanisms of approved compounds

Membraneless organelles (MLOs) formed by liquid-liquid phase separation contribute to intracellular compartmentalization of specific biological functions. The dynamic regulation of MLOs is critical for RNA metabolism, stress response, and signal transduction. While the number and size of MLOs serve as valuable indicators of cellular states, leveraging their spatial distribution and diverse quantitative characteristics to infer cellular conditions has posed significant challenges. Here, we present processing body (PB)-scope: an unsupervised deep learning-based framework designed for imaging-based phenotypic screening of PB, one representative MLO. The model was trained on a substantial dataset comprising over 400,000 single-cell confocal microscopy images from a colon cancer cell line treated with 280 compounds. PB-scope enabled precise drug classification based on their effects on PB characteristics, including number and size. This artificial intelligence framework offers insights previously obscured by the dynamic and subtle nature of PBs. Notably, PB-scope identified drugs with a shared function in the JAK signaling pathway, a previously unrecognized regulator of PB dynamics. PB-scope holds the potential for adapting to the study of other MLOs, offering broad applicability in this emerging field of cell biology.