Physics-informed neural networks (PINNs) for high-resolutional prediction of shear stress on cells in suspension culture

Abstract

The effect of shear stress on cell behaviors should be considered for designing the suspension culture of mammalian cells. Computational flow dynamics (CFD) is a promising tool for estimating shear stress on cells, but the accuracy is limited due to resolution limitations. In this research, we applied physics-informed neural networks (PINNs) for the high-resolution estimation of shear and drag stress on the cells in a swirling suspension culture. PINNs could complement the flow in the mesh and estimate the shear and drag stresses on the surface of cell particles smaller than the mesh size. The estimated shear and drag stress was lower than that from CFD calculation, and the shear stress depended on the non-dimensional number such as the Froude number. This approach could solve the limitation of the resolution of CFD for estimation of shear stress on the cells and is helpful to develop the large-scale suspension culture.