Predictive Modelling of Solvent Effects on Drug Incorporation into Polymeric Nanocarriers: A Machine Learning Approach.

This study aimed to identify solvent characteristics that enhance drug loading in polymeric micelles. Polyethylene glycol-block-polystyrene (PEG-b-PS) and curcumin were used as model compounds to investigate the impact of 40 different solvent mixtures on drug loading during flow-based assembly. We tested five algorithms: Random Forest (RF), Gradient Boosting (GP), XGBoost, Support Vector Regression (SVR), and Multilayer Perceptron (MLP), with the MLP model proving to be the most effective among them. To explain the model's predictions, we utilized SHapley Additive exPlanations (SHAP) values to identify solvent properties that contribute to high drug loading. Of the nine descriptors examined-curcumin solubility, polarity, Hildebrand solubility parameters, dipole moment, dielectric constants, viscosity, and Hansen solubility parameters (δD, δP, and δH)-solubility emerged as the most critical factor. Therefore, to achieve optimal drug loading, researchers should prioritize solvents with the highest solubility.