Integrating DLVO Theory with Machine Learning for Predictive Toxicology and Multi-Objective Design of Quercetin-Loaded Nanocarriers

The present study will develop a Physics-Informed Machine Learning framework for predicting cytotoxicity and optimizing the design of quercetin-loaded nanocarriers by integrating physical principles into data-driven modeling.

A dataset of 62 formulations curated from literature was characterized by physicochemical and biological descriptors. Multiple machine learning models, including Gradient Boosting and Support Vector Regression, were trained and validated. A PIML model featuring DLVO theory and kinetic constraints was developed. Paretooptimal formulations that balance cytotoxicity and loading efficiency were determined by multi-objective Bayesian optimization.

Gradient Boosting showed the best performance with R² = 0.91, while the predictions made by the PIML model were physically consistent, with R² = 0.88. Zeta potential, chitosan–alginate coatings, and sustained release behavior were identified as key determinants of biocompatibility.

This work shows the predictive power and interpretability of the PIML model for nanocarrier design, which enables safer and more efficient formulations of quercetin.