Abstract
Digital nucleic acid amplification tests enable absolute quantification of nucleic acids, but the generation of uniform compartments and reading of the fluorescence requires specialized instruments that are costly, limiting their widespread applications. Here, the authors report deep learning-enabled polydisperse emulsion-based digital loop-mediated isothermal amplification (deep-dLAMP) for label-free, low-cost nucleic acid quantification. deep-dLAMP performs LAMP reaction in polydisperse emulsions and uses a deep learning algorithm to segment and determine the occupancy status of each emulsion in images based on precipitated byproducts. The volume and occupancy data of the emulsions are then used to infer the nucleic acid concentration based on the Poisson distribution. deep-dLAMP can accurately predict the sizes and occupancy status of each emulsion and provide accurate measurements of nucleic acid concentrations with a limit of detection of 5.6 copies µl-1 and a dynamic range of 37.2 to 11000 copies µl-1 . In addition, deep-dLAMP shows robust performance under various parameters, such as the vortexing time and image qualities. Leveraging the state-of-the-art deep learning models, deep-dLAMP represents a significant advancement in digital nucleic acid tests by significantly reducing the instrument cost. We envision deep-dLAMP would be readily adopted by biomedical laboratories and be developed into a point-of-care digital nucleic acid test system.