CLASSIFICAÇÃO MULTIVARIADA PARA TRIAGEM CLÍNICA DE COVID-19 POR MEIO DA BIOESPECTROSCOPIA

Abstract

COVID-19, caused by the SARS-CoV-2 virus, is classified as a systemic disease and is primarily detected by serological and molecular methods that are laboratory-dependent. However, Fourier-transform infrared spectroscopy with attenuated total reflectance (ATR-FTIR) associated with chemometric methods has been studied for clinical screening of various diseases, including COVID-19, as it is a rapid and non-destructive technique that allows molecular-level information acquisition. Thus, the objective of this study was to evaluate different multivariate classification approaches in distinguishing between serum samples from individuals infected with COVID-19 and symptomatic individuals with a negative diagnosis. For this, 167 serum samples from symptomatic patients were used, including 76 negatives and 91 positives (UFES Ethics Committee 51803621.1.0000.5060). ATR-FTIR spectra were collected using a Bruker Alpha II spectrometer (Bruker) in absorbance mode. The data were pre-processed, divided into a training set (n=117) and an external test set (n=50), and evaluated using variable selection and multivariate classification methods (GA-LDA, PLS-DA, and PF-URF). The Fisher-weighted random forest method (PF-RF) demonstrated 85% sensitivity, 73.9% specificity, and 80% accuracy. The variables of interest were predominantly in the regions ~3500 cm^-1 to ~3000 cm^-1, ~3000 cm^-1 to ~2800 cm^-1, ~1700 cm^-1 to ~1600 cm^-1, ~1595 cm^-1 to ~1512 cm^-1, and ~1196 cm^-1 to ~1090 cm^-1, assigned to macromolecules of lipids, fatty acids, proteins, carbohydrates, and nucleic acids, respectively. This reinforces the applicability of ATR-FTIR of biofluids associated with multivariate classification for clinical disease screening.