Data-Driven Prediction of Nanoparticle Biodistribution from Physicochemical Descriptors

Nanoparticles have gained significant attention in biomedicine, electronics, and environmental science due to their unique physicochemical properties, which critically influence their absorption, distribution, metabolism, and excretion behavior in biological systems. However, predicting nanoparticle biodistribution and pharmacokinetics remains challenging due to the complexity of biological systems and the reliance on animal-derived data for physiologically based pharmacokinetic (PBPK) modeling. To address these limitations, this study integrates PBPK modeling with quantitative structure-activity (QSAR) relationship principles and multivariate linear regression (MLR) to develop a predictive framework for nanoparticle biodistribution based solely on physicochemical properties, using biodistribution data from healthy mice. Focusing exclusively on nondissolvable nanoparticles, we employed Bayesian analysis with Markov chain Monte Carlo simulations to fit PBPK models and generate kinetic parameters. The MLR-PBPK framework demonstrated strong predictive accuracy for kinetic indicators (adjusted <i>R</i>² up to 0.9) and successfully simulated nanoparticle biodistribution across 18 experiments. Key physicochemical properties such as zeta potential, size, and coating were identified as the most influential predictors, while the core material and shape had lesser impacts. Despite its success, the model faced limitations in predicting concentration-time curves for certain nanoparticles, highlighting the need for expanded data sets and nonlinear modeling approaches. This study provides a robust, nonanimal alternative for nanoparticle risk assessment, advancing safe and sustainable by design (SSbD) frameworks and offering a valuable tool for early-stage nanoparticle evaluation and design.