Researchers at the University of Osaka have deployed artificial intelligence to establish a unified framework for characterizing the microscopic structural order of supercooled water, a discovery published in Communications Chemistry. The study addresses a longstanding challenge in physical chemistry: while water’s anomalous behavior, such as its expansion upon freezing, is well documented, the absence of a systematic method to quantify how its molecular architecture evolves under varying temperature and pressure has hindered precise thermodynamic modeling. Water’s structural complexity arises from a dynamic network of intermolecular hydrogen bonds. As temperatures drop below the freezing point without solidification, the liquid enters a supercooled state where two competing microscopic configurations emerge: a high-density liquid with collapsed molecular arrangements and a low-density liquid characterized by open, tetrahedral geometries. Previous research has attempted to capture these transitions using independently developed structural descriptors, including tetrahedral bond order and local density metrics. However, the lack of a standardized evaluation method made it difficult to compare their predictive accuracy or determine which best reflects water’s underlying structural fluctuations. To resolve this, the Osaka team integrated a neural network trained on molecular dynamics simulation data. The model employed a trial-and-error learning process to analyze how well sixteen distinct descriptors could differentiate between high-density and low-density states across a range of subzero temperatures. By treating the classification task similarly to pattern recognition, the network isolated the most informative descriptors, effectively ranking them by their capacity to encode critical structural information. Corresponding author Kang Kim noted that incorporating machine learning allowed for a cognitive-like evaluation of descriptor efficacy, while senior author Nobuyuki Matubayasi emphasized that the network’s comparative analysis pinpointed the most efficient metrics for structural characterization. The resulting framework provides a standardized approach for evaluating and selecting structural descriptors, offering a clearer pathway to link microscopic molecular arrangements with macroscopic thermodynamic behavior. Researchers anticipate that this systematic methodology will deepen the understanding of water’s anomalous properties, improve computational models of supercooled phases, and guide the development of more precise descriptors for complex liquid systems. The findings underscore the growing utility of AI-driven pattern recognition in advancing fundamental physical chemistry and materials science.