Researchers at Princeton University have developed an artificial intelligence system capable of classifying biomolecular condensates into distinct morphological categories, revealing new insights into how pharmaceutical compounds influence cellular function. Published June 4 in the journal Cell, the study demonstrates how machine learning can map structural changes in vital intracellular droplets to specific drug-induced outcomes. Biomolecular condensates are membrane-less organelles that regulate transcription, RNA processing, and protein synthesis. Dysregulation of these structures has been associated with neurodegenerative diseases and cancer. Despite their importance, analyzing the dynamic shape shifts of these droplets across thousands of cells remains a significant bottleneck in traditional microscopy. Lead researcher Cliff Brangwynne, June K. Wu 92 Professor of Chemical and Biological Engineering at Princeton University, noted that the team needed a method to extract emergent structural patterns directly from complex image data. To address this, the researchers constructed a specialized neural network trained to analyze high-resolution microscope images of the nucleolus, a condensate responsible for ribosome assembly. After imaging hundreds of human cells under various drug treatments, the model categorized nucleolar morphology into four distinct classes. Three categories aligned with previously documented stress responses, while the fourth represented a previously unobserved structure. The AI successfully identified cap-shaped condensates following exposure to conventional anticancer agents, indicating a novel mechanism of action that disrupts RNA production pathways. Similarly, necklace morphologies emerged in response to compounds targeting separate RNA-related processes. The network also processed concentration gradients, quantifying dose-dependent shape transitions with high precision. Most notably, the system flagged a unique flower-like morphology induced by the cancer therapeutic topotecan. Further biochemical validation confirmed that topotecan inhibits the TOP1 enzyme, and this inhibition directly triggers the novel structural change. The discovery clarifies the enzyme role in maintaining nucleolar architecture through RNA processing regulation. Brangwynne emphasized that the neural network isolated this pattern precisely because it did not conform to established classification templates. Beyond nucleolar analysis, the team validated the framework on other RNA-associated condensates, including nuclear speckles and respiratory syncytial virus aggregates. The model consistently correlated specific morphological shifts with targeted drug concentrations, demonstrating broad applicability across cellular and viral systems. The study establishes a scalable, high-throughput methodology for evaluating single-cell drug responses. By translating complex visual data into quantifiable structural biomarkers, the AI tool enables researchers to detect subtle phenotypic changes that conventional size-based measurements typically miss. This capability accelerates drug discovery pipelines and offers a robust platform for monitoring therapeutic efficacy at the molecular level.