AUTOMATION OF FLUORESCENCE-ACTIVATED DROPLET RELEASE WORKFLOW BY DEEP LEARNING-BASED DROPLET DETECTOR

Water-in-oil droplets are becoming promising microreactors for many high-throughput biochemical applications.Compared with microwell plate-based assays, partitioning reagents to nL or pL droplets benefits from reduced sample consumption and reagent cost by 3-6 orders of magnitude per reaction.[1]However, droplets are still in limited use for specific scenarios where continuous monitoring of reaction kinetics and subsequent retrieval of targeted reactions are needed, such as directed evolution based on enzymatic kinetics.Though droplets can already be confined in microfluidic traps for continuous monitoring, the difficulty lies in the on-demand retrieval of targeted droplets from traps.To this end, we presented an on-demand droplet release method by laser-induced bubble employing the photothermal conversion capability of our previously developed active plasmonic fluorosurfactant (Figure 1) [2].The automation of fluorescence-activated droplet release (FADR) workflow is achieved with a custom-trained droplet detector based on YOLO (You Only Look Once) model [3].