Machine Learning-Enhanced Point of Care SERS Detection of SARS-CoV2 Variants Using Handheld Raman Spectrometer

The rapid evolution of SARS-CoV-2 and its emerging variants necessitates advanced diagnostic techniques for effective pandemic management. On-field detection of disease-causing pathogens is one of the primary health concerns. This study introduces a Machine Learning (ML)-enhanced Surface-Enhanced Raman Scattering (SERS) methodology for the precise differentiation of distinct SARS-CoV-2 strains and sub-strains in clinical samples. Ag nanorods (AgNR) were fabricated using physical vapor deposition method glancing angle deposition (GLAD) were used as highly pristine SERS substrates. Characterization of AgNR substrates were performed using FESEM, AFM, HRTEM, EDX, XRD, UV-Vis and contact angle. Our research targets the detection of four different strains of SARS-CoV2: Wildtype, Kappa, Delta, and Omicron, including their respective sub-strains (BA.1, BA.2, BA.5 and XBB). Using pristine AgNR arrays and a handheld Raman spectrometer, discernible spectral variations were observed. Despite the clarity in isolated cultured strains of viruses, clinical validation using nasopharyngeal swabs from positive samples presented complexities due to spectral overlaps. By harnessing the unique molecular vibrational patterns elucidated by Raman spectroscopy, SERS offers heightened sensitivity. But problems came up because of the small differences in spectral patterns between closely related SARS-CoV-2 variants found in clinical samples. To address this, Machine learning (ML) algorithms were integrated to discern intricate patterns from SERS data, enhancing differentiation capabilities. Through the integration of gradient boost (GB) and support vector machine (SVM) models of ML within the SERS framework, our approach achieved an accuracy of 89% and 94% respectively in identifying targeted variants from nasal swabs of human patients. This integrated ML-SERS approach not only enhances detection efficacy but also offers cost-effective on-site detection capabilities and also disease prediction ability. The demonstrated precision underscores the methodology's potential in future variant identification and pandemic surveillance.