A hybrid approach to aqueous solubility prediction using COSMO-RS and machine learning

Abstract

Solubility prediction plays a pivotal role across pharmaceutical development, from drug discovery through to process optimisation. This study presents a hybrid approach that leverages thermodynamic models and machine learning to develop a solubility model. The COSMO-RS theoretical framework, implemented via the COSMOtherm software, produced conformer specific features for the machine learning model. This feature extraction methodology achieved high predictive power while requiring only a fraction of the data compared to traditional machine learning methods. This framework empowers predictive aqueous solubility modelling without the necessity for solute-specific experimental data, a critical advantage in the early phases of drug discovery.