AI‐Powered, Temperature‐Resolved Centrifugal Microfluidics for Rapid 3D Phase‐Diagram Generation of Biomolecular Condensates

Biomolecular phase separation drives the formation of membraneless organelles and is implicated in numerous diseases. Mapping multidimensional phase diagrams is essential for understanding condensate assembly mechanisms and developing therapeutic interventions, yet current methods are slow, resource-intensive, and often limited to a single, frequently non-physiological, temperature. Here, we present T-PhaseMap, an AI-powered centrifugal microfluidic platform that integrates programmable concentration-gradient generation, standalone temperature control, and deep-learning analysis to construct 3D phase diagrams across composition-temperature space within 30 min. We demonstrate a dual-scale strategy that enables broad-range exploration of phase space and localized refinement of critical transition boundaries. Meanwhile, tunable tri-nodal temperature control at 25°C, 37°C, and 45°C provides a structured thermal dimension for systematically evaluating temperature-dependent phase behavior. A VGG-16-based four-class classifier further distinguishes two-phase condensates, single-phase states, boundary states, and aggregate-like phenotypes, achieving approximately 99.3% accuracy on an independent test set of 720 images. Applied to heparin sodium-mediated condensate modulation, T-PhaseMap revealed a 17.6% temperature-dependent variation in apparent suppression potency across 25°C-45°C. This platform provides a rapid and objective tool for physiologically relevant phase-diagram generation and high-throughput screening of condensate-targeting therapeutics.