Artificial intelligence-assisted physiologically based pharmacokinetic modeling and large-scale machine learning analysis of tumor delivery of nanoparticles in mice

Nanomedicine is a promising strategy for cancer therapy. However, the low delivery efficiency of nanoparticles (NPs) to tumors remains a challenge and hinders clinical translation. In this study, an artificial intelligence (AI)-assisted physiologically based pharmacokinetic (PBPK) model was developed to predict NPs' tumor delivery efficiency (DE) based on the physicochemical properties and experimental strategies. First, a generic PBPK model was applied to describe NPs’ tumor kinetics, and six different tumor-related parameters were determined. Second, machine learning (ML) and AI models were developed for predicting the tumor-related parameters. Incorporating the ML-predicted parameters into the PBPK model, an ML-assisted PBPK model was developed to predict the tumor kinetics of NPs. This generic PBPK model was applied to simulate the tumor kinetics, in which 114 out of 152 datasets (75%) achieved the determination coefficient (R 2 ) of ≥0.75, and the overall R 2 was 0.92. Distribution (PT) and permeability (PATC) coefficients at the tumor site were the most important parameters in predicting tumor DEs. The deep neural network (DNN) model was the best-performing model and selected for tumor-related parameter estimations. Targeting strategy and surface charge were determined as the important features for predicting various tumor-related PBPK parameters. External validation was conducted using an in vivo experiment, and half of the predicted concentrations were within the two-fold error of the observed data. This model can serve as a complementary tool for predicting tumor kinetics prior to animal experimentation, helping identify promising NP formulations and potentially improving the success rate of drug delivery.