Prediction of Drug Solubility in Complex Micro-Environments: Machine Learning Study Based on Hansen Solubility Parameters

Although machine learning (ML) has been widely applied to drug solubility prediction, most existing models focus on single-solvent systems, and accurate prediction in complex micro-environments (mixed solvents) remains challenging. Herein, we report the first predictive framework that integrates mechanism-driven microscopic variables (molecular descriptors and Hansen solubility parameters) into three ML algorithms: artificial neural networks (ANN), support vector machines (SVM), and random forests (RF). Among them, RF achieved the best performance, with the coefficient of determination (R2) markedly improved from 0.830 to 0.988 when Hansen parameters were included as input features. Analysis of factor interactions revealed that solvent hydrogen-bonding capacity, polarity, mixing ratio, and temperature play key roles in modulating drug solubility, consistent with previous experimental studies. These results underscore the critical role of microscopic variables in capturing solubility behavior in mixed solvent systems. More broadly, this work demonstrates that integrating mechanism-driven descriptors with data-driven ML models offers a powerful and generalizable strategy for accurately predicting drug solubility in complex micro-environments.