Machine Learning-driven Optimization of Doxorubicin-loaded Nanocomposites: Predicting and Enhancing Loading and Encapsulation Efficiencies

Background The therapeutic effect of doxorubicin (DOX) chemotherapy is limited by its severe systemic toxicities, which are minimized by nanocomposite-based drug delivery systems to maximize Loading Efficiency (LE%) and Encapsulation Efficiency (EE%). Empirical method-based traditional optimization proves ineffective and fails to capture complex parameter interactions. Methods In the current study, an Machine learning (ML) model was established to forecast and optimize LE% and EE% of DOX-loaded nanocomposites. A duly obtained list of 77 different formulations from literature was used to train and validate multiple ML models, i.e., Random Forest (RF), Gradient Boosting (GB), and XGBoost. The top-performing model was interpreted by SHapley Additive exPlanations (SHAP) analysis, and multi-objective optimization was performed using the Non-dominated Sorting Genetic Algorithm II (NSGA-II) algorithm. Results For LE% and EE%, the RF model fared better with R² values of 0.86 and 0.83, respectively. Nanocarrier size, zeta potential, chitosan, and hyaluronic acid were the most important features. Multi-objective optimization identified best design parameters, proposing chitosan-based nanocomposites of sizes 100–200 nm and moderately negative zeta potentials (-20 to -30 mV) for simultaneous maximization of both efficiency indicators. Conclusion The research presents a holistic, data-driven method to rational nanocarrier design that significantly accelerated the development of advanced DOX delivery systems through the replacement of time-consuming trial-and-error efforts with a computer system.