Machine Learning-Assisted,\nDual-Channel CRISPR/Cas12a\nBiosensor-In-Microdroplet for Amplification-Free Nucleic Acid Detection\nfor Food Authenticity Testing

The development of novel detection technology for meat\nspecies\nauthenticity is imperative. Here, we developed a machine learning-supported,\ndual-channel biosensor-in-microdroplet platform for meat species authenticity\ndetection named CC-drop (<u>C</u>RISPR/Cas12a digital\nsingle-molecule microdroplet biosensor). This strategy allowed us\nto quickly identify and analyze animal-derived components in foods.\nThis biosensor was enabled by CRISPR/Cas12a-based fluorescence lighting-up\ndetection, and the nucleic acid signals can be recognized by a Cas12a–crRNA\nbinary complex to trigger the <i>trans</i>-cleavage of any\nby-stander reporter single-stranded (ss) DNA, in which nucleic acid\nsignals can be converted and amplified to fluorescent readouts. The\nultralocalized microdroplet reactor was constructed by reducing the\nreaction volume from up to picoliter to accommodate the aforementioned\nreaction to further enhance the sensitivity to even render an amplification-free\nnucleic acid detection. Moreover, we established a smartphone App\ncoupled with a random forest machine learning model based on parameters\nsuch as area, fluorescent intensity, and counting number to ensure\nthe accuracy of image recording and processing. The sample-to-result\ntime was within 80 min. Importantly, the proposed biosensor was able\nto accurately detect the <i>ND1</i> (pork-specific) and <i>IL-2</i> (duck-specific) genes in deep processed meat-derived\nfoods that usually had truncated DNA, and the results were more robust\nand practical than conventional real-time polymerase chain reaction\nafter a side-by-side comparison. All in all, the proposed biosensor\ncan be expected to be used for rapid food authenticity and other nucleic\nacid detections in the future.