Samsung Advanced Institute of Technology

Predicting both accurate and reliable solubility value has long been a crucial but challenging task. In this work, novel methods have been developed to accurately predict the solubility of two molecules (solute and solvent) through machine learning and deep learning. The current study employed two methods: 1) molecules are converted into molecular fingerprint and optimal physicochemical properties were added as descriptors 2) graph convolutional network (GCN) models were used to convert molecules into graph representation and deal with prediction tasks. Then, prediction tasks were conducted for each method: 1) the solubility value (regression) and 2) the class (classification). The fingerprint-based method clearly demonstrates that achieving high performance is possible by adding simple but significant physicochemical descriptors to molecular fingerprints. Meanwhile, the GCN method have shown that reaching remarkable results in predicting various properties of chemical compounds is possible only with relatively simplified features from the graph representation. The developed methodologies provide comprehensive view of constructing a proper model for predicting solubility and can be employed for finding the suitable solute and solvent.