Label-free surface-enhanced Raman spectroscopy and machine learning for rapid classification and quantification of...

Bovine respiratory disease complex (BRDC) is one of the most economically significant multifactorial diseases affecting the global cattle industry, creating an urgent need for rapid, sensitive, and field-deployable diagnostic technologies. Here, we report a label-free surface-enhanced Raman spectroscopy (SERS) platform integrated with machine learning (ML) for the simultaneous classification and quantification of five major BRDC viruses: bovine respiratory syncytial virus (BRSV), bovine viral diarrhea virus types 1 and 2 (BVDV-1 and BVDV-2), infectious bovine rhinotracheitis virus (IBRV), and bovine parainfluenza virus type 3 (BPIV-3). Reproducible SERS spectra were acquired from serially diluted virus specimens in deionized water using silica-coated silver nanorod substrates fabricated by oblique angle deposition, while spectra of deionized water and Dulbecco's Modified Eagle Medium (DMEM) were collected separately as background controls. Despite shared biochemical components, distinct virus-specific spectral fingerprints were observed and analyzed using supervised machine learning. SVM classification achieved an overall accuracy of 99.57% with no misclassification between viral and background spectra. SVR enabled accurate viral concentration prediction over multiple orders of magnitude, yielding an overall R² of 0.9974, a MAE of 0.028, and a RMSE of 0.0373. For blind specimens measured at concentrations excluded from model training, virus identification based on majority-vote SVM achieved 100% accuracy, while SVR-based regression retained a concentration-dependent prediction trend for withheld dilution levels. Overall, these results establish a label-free SERS-ML framework for rapid BRDC virus identification and quantitative concentration prediction under controlled conditions, providing a proof-of-concept foundation for future veterinary diagnostic development.