ATR-FTIR spectroscopy and chemometric analysis as highly sensitive detection tool for SARS-CoV-2 infection in saliva from different types of COVID-19 severity

Rapid, non-invasive, and scalable diagnostic tools are essential for managing respiratory viral infections and enhancing pandemic preparedness. While vibrational spectroscopy coupled with chemometrics has emerged as a powerful biosensing tool, many SARS-CoV-2 studies remain constrained by limited sample sizes and narrow clinical characterization. This study evaluates the performance of attenuated total reflection Fourier-transform infrared (ATR-FTIR) spectroscopy for detecting SARS-CoV-2 in saliva samples from a heterogeneous cohort of 207 symptomatic individuals with varying clinical severities. Using RT-qPCR as the gold standard, saliva spectra were processed using multivariate classification and variable selection algorithms. The optimized model, based on second-derivative Savitzky-Golay preprocessing and interval Successive Projections Algorithm-Partial Least Squares Discriminant Analysis (2d-SG/iSPA-PLS-DA), achieved an overall accuracy of 91.3 %, with 93 % sensitivity and 87 % specificity. These findings underscore that ATR-FTIR spectroscopy of saliva, supported by robust variable selection strategies, provides a high-performance, reagent-free screening platform suitable for clinical triage and large-scale surveillance of respiratory pathogens.