Neural Surrogate ODEs for Accelerated PK/PD Modeling and Clinical Trial Simulation of Anti-Cancer Drugs

Background Anticancer drug development is limited by the low computational efficiency of traditional pharmacokinetic/pharmacodynamic (PK/PD) modeling in large-scale clinical trial simulations, hindering rapid optimization of individualized dosing regimens and subpopulation analyses, especially for rare cancers with scarce data. Methods We developed a Neural Surrogate ODEs framework that replaces conventional PK/PD solvers with a neural surrogate to enable efficient and mechanistically consistent simulations. The framework compresses high-dimensional PK/PD dynamics, including drug concentration and tumor response trajectories, into a low-dimensional latent space. A hybrid regularization strategy enforces key pharmacological principles while preserving data-driven flexibility. The model was validated using PK/PD data from three established anticancer drugs (paclitaxel, imatinib, and sunitinib) and one rare cancer therapy (vemurafenib for melanoma). Results We developed a Neural Surrogate ODEs framework that replaces conventional PK/PD solvers with a neural surrogate to enable efficient and mechanistically consistent simulations. The framework compresses high-dimensional PK/PD dynamics, including drug concentration and tumor response trajectories, into a low-dimensional latent space. A hybrid regularization strategy enforces key pharmacological principles while preserving data-driven flexibility. The model was validated using PK/PD data from three established anticancer drugs (paclitaxel, imatinib, and sunitinib) and one rare cancer therapy (vemurafenib for melanoma). Conclusions This study integrates PK/PD modeling with artificial intelligence, providing a scalable computational tool for oncology drug development. The combination of mechanistic modeling and neural surrogate technology offers a novel approach with direct implications for personalized cancer therapy and rare cancer drug development.