Surface-enhanced raman spectroscopy with machine learning in non-invasive detection of dengue-NS1 fingerprint

Abstract

The surface-enhanced Raman spectroscopy (SERS) method exploits the plasmonic effect of nano-sized metallic materials to intensify the Raman scattering of the monochromatic light of analyte molecules. This promotes the sensitivity and specificity of the Raman spectroscopy analysis method. This study integrated SERS with machine learning (ML) to detect dengue fever, a disease infecting more than 40% of the world’s population. Non-structural protein 1 (NS1), detected in the sera of infected dengue patients during the early infection stage, is currently recognised as a biomarker for the early diagnosis of DF. However, no attempts have been made to detect NS1 in the salivary Raman spectra. Given this situation, this study delves into the potential of SERS as an early, non-invasive DF detection technique using salivary NS1. The SERS spectra of saliva samples (n=289) were collected and subsequently classified as positive and negative for DF, using principal component analysis (PCA) integrated with support vector machine (SVM) models. The PCA-SVM model's performance was benchmarked against two clinical diagnostic NS1-enzyme-linked immunosorbent assay (ELISA) tests recommended by the World Health Organization (WHO). The PCA-SVM model outperformed both tests regarding radial basis function kernel (RBF) and cumulative percent variance (CPV; 83.22% accuracy, 88.27% sensitivity, 78.13% specificity). It is encouraging that the sensitivity level of the PCA-SVM model is above the benchmark set by the saliva-based NS1-ELISA tests proposed by the WHO, demonstrating the potential of SERS for the non-invasive detection of DF.