A physics-informed neural network framework for quantitative analysis of transcytosis and physical diffusion in an in vitro BBB

Nanoparticles traverse the blood–brain barrier (BBB) through passive diffusion and vesicular transcytosis, but the quantitative contributions of these routes remain difficult to determine. Here, we combine a controlled in-vitro human BBB model (hCMEC/D3 Transwells) with a physics-informed neural network (PINN) to interpret transport kinetics and estimate paracellular and vesicular components. Monodisperse polystyrene nanoparticles (20, 50 and 120 nm) showed low polydispersity, stable ζ-potential and minimal cytotoxicity. Intact monolayers displayed high TEER and low tracer permeability, whereas TNF-α induced reversible junctional opening. Apical-to-basolateral transport increased with junctional loosening and remained size-dependent; clathrin and dynamin inhibition reduced flux without altering TEER or tracer passage. A mass-balance-constrained PINN incorporating a TEER-linked permeability term reproduced transport profiles and generalized to combined perturbation (TNF-α + chlorpromazine). Under our conditions, the model suggested that vesicular uptake represented the major route, with a smaller diffusion component that increased during junctional disruption and clathrin inhibition. Overall, this combined experimental–computational approach provides a practical framework for pathway-informed evaluation of nanoparticle transport across the BBB.