Abstract
Microfluidic affinity-based capture of extracellular vesicles (EVs) holds great promise for disease diagnostics and monitoring therapeutic EV production. However, optimizing microfluidic channel geometries for efficient EV capture remains understudied due to the effort required to experimentally test numerous designs. To overcome this, we developed an automated parallel pattern search (PPS) optimizer integrating Python, COMSOL Multiphysics, and high-performance computing. We applied this approach to optimize triangular micropillar array geometries by parameterizing periodic unit cells and maximizing simulated particle capture efficiency. Surprisingly, the highest EV capture was not achieved by maximizing the number of pillars and surface area. Instead, designs with slightly larger, more widely spaced pillars promoted EV contact by enabling slower particles to follow the contours of the pillars more closely. We validated these findings experimentally using bioreactor-produced EVs and microchannels functionalized with anti-CD63 antibodies. EVs captured in the best and worst designs were quantified using fluorescence plate reading and eluted for nanoparticle tracking analysis (NTA), confirming a significant increase in capture efficiency with the optimized design. These results highlight the power of automated microfluidic optimization to advance EV isolation technologies and demonstrate a practical strategy for improving device performance in this rapidly growing field.