Model-guided design of lymphatic drug delivery systems using osmotic pressure, viscosity and lymph node size.

Metastatic lymph nodes (LNs) represent a major barrier to effective drug delivery in cancer therapy. To address this challenge, a lymphatic drug delivery system (LDDS) enabling direct intranodal administration has been developed. In LDDS, drug transport and retention are governed by formulation osmotic pressure and viscosity; however, optimization strategies according to LN size and therapeutic objectives remain unclear. Here, we established a quantitative framework that integrates mathematical modeling with machine and deep learning to describe and predict intranodal drug kinetics as a function of osmotic pressure, viscosity and Maximum cross-sectional area of LN (MALN). Using a unique mouse model possessing human-sized LNs, we first constructed a parametric model that accurately reproduced fluorescence intensity profiles in LNs. The model captured key kinetic features and is enabled to apply time-series datasets. A multilayer perceptron-based surrogate model was then trained to predict C(t) profiles and pharmacokinetic parameters across the entire π-μ-MALN space. In parallel, downstream LN transport exhibited two distinct Tmax phenotypes, which were successfully classified using machine learning to evaluate network-level delivery efficiency. Importantly, comparison with clinical LDDS data demonstrated that the model also describes early drug kinetics in human LNs. Because this framework relies only on clinically accessible LN size information, it enables inverse design of osmotic pressure and viscosity tailored to LN size and desired delivery outcomes. This study provides a mechanistically grounded and translatable strategy for rational LDDS formulation design to support controlled lymphatic drug delivery in future clinical applications.