Deep Learning-Enabled Real-Time Single-Shot Refocusing of Microwell Array for Digital Melting Curve Analysis

Digital melting curve analysis (dMCA) represents a breakthrough technology for multiplexed nucleic acid detection within limited fluorescence channels, utilizing thermal melting imaging postdigital PCR. However, conventional dMCA suffers from accuracy degradation across broad temperature ranges and resolution limitations due to defocusing-induced fluorescence deviations. To overcome these constraints, we propose a novel deep learning-enabled dMCA platform based on a single-shot adaptive point-spread function (PSF) attention refocusing model (SAPAR-dMCA), autofocusing the digital PCR microarray without electromechanical motion. Our platform achieves a ±400 μm depth-of-field via PSF self-calibrating and modulation, reducing fluorescence intensity deviation within microwells by 2.76-fold compared with state-of-the-art methods. Experimental validation demonstrates significantly enhanced multiplex identification accuracy (from 38.0 to 92.3%) across a 46.0 °C melting temperature span, with the coefficient of variation reduced from 3.16 to 0.78%. Applied to respiratory pathogen detection, SAPAR-dMCA attains 0.9 °C resolution based on melting temperature differentiation. This work establishes SAPAR-dMCA as a precise and robust platform for digital multiplex nucleic acid analysis, featuring unprecedented temperature range adaptability and high resolution. Our methodology paves the way for ultramultiplexed gene profiling and precision medicine advancement.