This project seeks to transform our understanding of how liquids, such as water and complex solutions, behave at the molecular level by developing innovative computational tools based on the emerging computer science field of graph representation learning. These tools will treat liquids as dynamic networks, where molecules and their interactions are represented as nodes and edges, enabling unprecedented insights into their structure, dynamics, and properties. This approach addresses a critical gap in current methods, which struggle to predict liquid behavior accurately, especially in complex or extreme environments. By uncovering the connections between molecular interactions and large-scale properties, this research will advance fields ranging from energy storage to drug design to environmental science. The project’s integration of state-of-the-art algorithms with experimental validation has the potential to accelerate the discovery of new materials and processes, benefiting biochemical and technological innovation. This project aims to develop a novel framework combining graph representation learning (GRL) with molecular simulations and experiments to characterize and predict the behavior of liquids and solutions at the molecular level. By representing liquids as spatiotemporal graphs, where molecules and their interactions are nodes and edges, the project seeks to identify structural and dynamic patterns that connect intermolecular interactions to macroscopic properties