Abstract: 

Solubility represents a fundamental parameter with wide-ranging applications across diverse scientific domains, including pharmaceuticals, renewable energy, and electrochemistry. Nevertheless, the experimental determination of solubility faces substantial challenges, particularly in scenarios involving ever-expanding solution combinations. In light of these experimental challenges, computational methods have gained prominence as viable means of solubility prediction. In this context, this seminar will discuss two research papers that have made notable contributions to the field of solubility prediction, specifically through the utilization of graph neural networks. These papers offer significant achievements in addressing critical issues related to data scarcity and the incorporation of chemical insights. Building on the motivations gained from these studies, the seminar also outlines ongoing and prospective research of mine, aimed at enhancing chemical space coverage of data and expanding the model to be applicable in multicomponent solvation systems.