Decoding the impact of administration route and nanocarrier composition on systemic pharmacokinetics of drug-loaded nanoparticles

Accurate prediction of nanoparticle-mediated pharmacokinetics remains challenging due to the nonlinear interplay between formulation properties and biological systems. This study aims to develop a hybrid machine learning framework integrating ensemble models and metaheuristic optimization to predict key pharmacokinetic endpoints (Cmax, Tmax, and AUC). This study, based on a dataset of 403 samples from the literature comprising physicochemical and biological descriptors, coupled XGBR-LIM and ADAR-LIM ensembles with two contrasting optimizers (the exploratory SHO and the conservative SaBO). The hybrid models deliver high predictive power ( R 2 up to 0.93, CCC > 0.88, with narrow prediction interval widths ranging from 0.06 to 0.10 for the best-performing hybrid models), with lower RMSE for steps like AUC, which are smoother, and greater variability for absorption-driven Tmax. PIB and LIME analysis uncover substantial and consistent contributions from nanoparticle size, zeta potential, entrapment efficiency, and administration route. The results demonstrate that optimized hybrid models significantly improve predictive accuracy and stability across heterogeneous datasets, with uncertainty quantification further enhancing reliability.