A team at the University of Osaka has used artificial intelligence to uncover hidden molecular structures behind water's unusual behavior, such as expanding when it freezes. Their study, published in Communications Chemistry, introduces an AI-based framework that evaluates and compares different ways of describing the structure of supercooled water, which is water that remains liquid below its normal freezing temperature when nucleation sites are absent. One leading explanation suggests supercooled water shifts between two competing liquid structures: a high-density liquid (HDL) and a low-density liquid (LDL), created by an ever-changing network of hydrogen bonds. Researchers have developed many structural descriptors, such as tetrahedral bond order and local density, but they rely on different scales, dimensions, and types of information, making direct comparison difficult. Corresponding author Kang Kim stated that past studies showed machine learning is effective for classifying structural data, and the team specifically wanted to incorporate a neural network model to evaluate how accurate the descriptors were at capturing key structural information.