Towards the Prediction of Drug Solubility in Binary Solvent Mixtures at Various Temperatures Using Machine Learning

Abstract Drug solubility plays an important role in the drug development process. Traditional methods for measuring solubility involve saturating a solvent with the drug and determining the drug concentration thereafter. However, these techniques are tedious and challenging to employ when dealing with expensive drugs or those available in small quantities. To address this, researchers have begun to leverage machine learning (ML) as an alternative approach. ML offers a data-driven strategy that enables the training of models on existing datasets to predict unmeasured solubility. Nonetheless, the majority of existing ML research has focused on the predictions of aqueous solubility and/or solubility at specific temperatures. This restricts the application of these models in pharmaceutical development which often requires insights into drug solubility across various solvents, solvent mixtures, and temperature conditions. To bridge this gap, we compiled an extensive dataset including solubility of small molecules measured in a range of binary solvent mixtures under various temperatures. We trained models on this dataset and subsequently optimized through Bayesian optimization to identify the models and model configurations that deliver optimal performance. The chosen top-performing models were further validated through a prospective study. The results demonstrated the potential of these developed ML models to predict drug solubility, especially for drugs whose features closely align with the small molecules within the dataset. To support future research and facilitate advancements in the field, we have made the dataset and the codes openly available.