Analysis of Pharmacokinetic-Pharmacodynamic Relationships of Nanoparticles against Tumors

Nanoparticle (NP)-based drug delivery systems hold great promise for cancer treatment. However, designing efficient NP formulations for clinical usage remains a challenge. This study created a "Nano-PKPD Database" by curating pharmacokinetic (PK) data on NP tumor delivery and tissue biodistribution, as well as pharmacodynamic (PD) data on tumor volume changes in tumor-bearing mice. Various machine learning (ML) models were developed to explore the PK-PD relationship and predict antitumor efficacy based on NP physicochemical properties, experimental strategies, and PK metrics. The current database contains 611 data sets from 345 papers on NP time-dependent concentrations in tumors and major organs. The median delivery efficiency was 0.70 percentage of injected dose (%ID) in tumors, 0.17%ID (heart), 10.72%ID (liver), 0.59%ID (spleen), 0.33%ID (lung), and 0.96%ID (kidney). In addition, 833 data sets from 340 papers on time-dependent tumor volume changes were collected, where the median tumor growth inhibition was 63.12%. A total of 18 ML models were developed, where tree-based models achieved the best discriminative performance. The use of assistive technology, zeta potential, and targeting strategy were the top 3 features related to antitumor efficacy. This study reports an open-access database and multiple ML models for PK-PD investigation, facilitating nanomedicine design and accelerating clinical translation.