Rapid preliminary screening of Tomato brown rugose fruit virus based on surface-enhanced Raman spectroscopy and machine learning.

Introduction: Tomato brown rugose fruit virus (ToBRFV) represents a growing threat to global tomato production, causing severe losses in crop yield and fruit quality. Although the standard RT-qPCR assay is highly accurate, its reliance on laboratory processing, specialized equipment, and trained personnel limits its applicability for rapid on-site diagnostics. To address this limitation, this study evaluated a biosensing method that does not require labels and combines surface-enhanced Raman scattering (SERS) with machine learning to distinguish tomato leaves infected with ToBRFV from healthy leaves. Methods: Following health status confirmation via RT-qPCR, leaf extracts were directly deposited onto silver nanorod arrays for SERS spectral acquisition. Three classification models were evaluated: PCA-LDA, PLS-DA, and SVM. Results: The results showed that all models were able to discriminate infected samples from healthy samples in the present dataset. Notably, the SVM model exhibited the best performance, achieving an accuracy of 91.67%, a sensitivity of 100.00%, a specificity of 81.48%, and an area under the ROC curve (AUC) of 0.993. Discussion: This result suggests that SERS spectra may contain biochemical information associated with ToBRFV infection and that such information can be used for sample classification using machine learning models. In its present form, this approach is intended as a rapid, low-cost, field-deployable preliminary screening tool - not a replacement for RT-qPCR or other confirmatory molecular assays. The reported accuracy was obtained on mechanically inoculated plants of a single cultivar under controlled greenhouse conditions and should therefore be interpreted as a proof-of-concept upper bound; field-scale validation is the focus of ongoing work.