Machine learning and Raman spectroscopy: A novel approach to cell research and early detection of citrus black spot

Citrus Black Spot (CBS), caused by Phyllosticta citricarpa , is characterized by distinct lesions, including hard spots, water-soaked lesions, and necrotic spots that develop throughout fruit growth. These symptoms lead to peel pitting, deformities, and premature fruit drop. Understanding the infection process and host defense mechanisms at the cellular micro-level, along with early disease detection, can provide new strategies for controlling CBS. We developed an integrated approach combining Raman spectroscopy, chemometrics, and machine learning for cellular-level research and early CBS detection. A confocal micro-Raman spectrometer was utilized to obtain spectra from healthy and CBS-infected citrus at juvenile and mature stages, supplemented with Raman microspectroscopic imaging to create image cubes. Spectra from juvenile and mature citrus were preprocessed using Adaptive Iteratively Reweighted Penalized Least Squares (airPLS) and Standard Normal Variate (SNV), respectively. Comparison with reference spectra of citrus secondary metabolites identified carotenoids as biomarkers for CBS pathogenesis. By monitoring carotenoid Raman peak intensities and pseudocolor imaging, this study revealed for the first time at the cellular level that citrus modulates carotenoid metabolism as a defense response to P. citricarpa infection. The significant negative correlations between characteristic peak intensities and disease severity indices (r = -0.76, -0.89, -0.93; p <0.01) further supported this mechanism. Finally, Backpropagation (BP) neural network based on Principal Component Analysis (PCA) and Competitive Adaptive Reweighted Sampling (CARS) feature extraction algorithms demonstrated robust performance in the early detection of CBS. The CARS-BP achieved an accuracy of 97.83% for juvenile citrus, while the PCA-BP reached 97.59% accuracy for mature citrus.