Learning-aided design of micropost arrays for optimizing interface stability and mass transport in organs-on-chips

Abstract

Learning-aided design of micropostMicropost-based organ-on-a-chip (OOC) platforms are widely used to spatially confine cell-laden hydrogels and create three-dimensional (3D) co-culture microenvironments while maintaining adjacent channels for mass transport, drug delivery, and molecular sampling. During hydrogel injection, surface tension across the micropost array keeps the hydrogel interface stable and creates a semi-closed configuration that allows diffusion-based compounds exchange. Despite their widespread use, rational design of micropost geometries that simultaneously ensure interface stability without leakage and efficient mass transport remains largely empirical and lacks systematic quantitative guidelines. Here, we present a learning-aided computational framework for micropost design by performing high-throughput simulations of 160 distinct micropost geometries and integrating the results into a multi-objective Bayesian optimization scheme. Two-phase flow simulations were used to evaluate interface stability during injection, while diffusion simulations quantified transport efficiency across stably formed interfaces. By modeling the coupled relationships between geometric parameters and performance metrics, we identified optimal micropost designs that satisfy predefined stability and transport criteria. This framework provides a robust and extensible approach for rational micropost design and is readily applicable under varying flow conditions, material properties, or device dimensions, offering generalizable design guidance for micropost-based OOC systems.