Lyophilized nasal swabs for COVID-19 detection by ATR-FTIR spectroscopy: Machine learning-based approach

Abstract

The COVID-19 pandemic continues to pose challenges for global health. The disease burden and diagnostic pressure has forced scientists to explore alternate diagnostic tools beyond the standard PCR testing. One such promising tool is the use of spectroscopy-based diagnostics. The objective of this study is to assess the potential of ATR-FTIR spectroscopy, applied to lyophilized nasal swab samples to discriminate between healthy and infected COVID-19 patients. Equal number (55 each) of positive and negative freeze-dried nasal swab samples were analyzed. After pre-processing, average mean spectra (600–4000 cm⁻¹) showed significant variations between healthy and infected sample types. Clear spectral variations were recorded at 17 locations, of which, 13 peaks were observed in COVID-19 spectra while 4 peaks were observed in negative sample spectra. Statistical discrimination was done using principal component analysis (PCA), linear discriminant analysis (LDA) and support vector machine (SVM). The first two principal components (PCs) showed a combined variance of 76 %. Classification accuracy of 100 % were observed in the LDA graph using Quadratic kernel. Similarly, SVM model with both internal validation and external validation confirmed the robustness with a 100 % classification accuracy. These results show that lyophilized nasal swab samples are the ideal sample choice for FTIR-based analysis of COVID-19. This sample preparation method coupled with spectroscopy can serve as a robust and accessible diagnostic tool for post-covid testing.