Microfluidic liquid biopsies using affinity-based capture of extracellular vesicles (EVs) have demonstrated great potential for providing rapid disease diagnosis and monitoring. However, little effort has been devoted to optimising the geometry of the microfluidic channels for maximum EV capture due to the inherent challenges of physically testing many geometric designs. To address this, we developed an automated parallel pattern search (PPS) optimiser by combining a Python optimiser, COMSOL Multiphysics, and high performance computing. This unique approach was applied to a triangular micropillar array geometry by parameterising repeating unit cells, making several assumptions, and optimising for maximum particle capture efficiency. We successfully optimised the triangular pillar arrays and surprisingly found that simply maximising the total number of pillars and effective surface area did not result in maximum EV capture, as devices with slightly larger pillars and more spacing between pillars allowed contact with slower moving EVs that followed the pillar contours more closely. We then experimentally validated this finding using bioreactor-produced EVs in the best and worst channel designs that were functionalised with an antibody against CD63. Captured EVs were quantified using a fluorescent plate reader, followed by an established elution method and nanoparticle tracking analysis. These results demonstrate the power of automated microfluidic geometry optimisations for EV liquid biopsies and will support further development of this rapidly growing field.
On the Convergence of Asynchronous Parallel Pattern Search
