0
0The paper details the development of Deep-Phase, a deep-learning framework designed to interpret the complex morphology of biomolecular condensates to reveal their internal biochemical states. By analyzing microscopy images of the nucleolus, nuclear speckles, and viral inclusion bodies, the researchers demonstrated that structural changes can quantitatively predict drug potency and functional disruptions in RNA processes. This unbiased approach surpasses traditional image analysis by identifying morphological fingerprints without the need for manual feature selection. Notably, the authors utilized Deep-Phase to discover a novel "flower" nucleolar phenotype, uncovering a previously unknown role for the enzyme TOP1 in maintaining nucleolar organization and ribosomal RNA processing. Ultimately, the study presents a powerful platform for high-content screening and for understanding how mesoscale cellular structures relate to molecular pathways and therapeutic targets.
References:
Donlic A, Comi TJ, Quinodoz SA, Jaberi-Lashkari N, Antunes Fernandes K, Jiang L, Wiesner LW, Lim AI, Brangwynne CP. Deep learning of functional perturbations from condensate morphology. Cell. 2026 Jun 4:S0092-8674(26)00569-6. doi: 10.1016/j.cell.2026.05.010. Epub ahead of print. PMID: 42242225.
