Detection of People Positive to COVID-19 through ATR-FTIR Spectra Analysis of Saliva using Machine Learning

Abstract

COVID-19 is an infectious disease caused by the SARS-CoV-2 virus. This virus's spread is mainly through droplets released from the nose or mouth of an infected person. Although vaccines have been developed that effectively reduce the effects that this viral infection causes, the most effective method to contain the virus’s spread is numerous tests to detect and isolate possible carriers. However, the response time, combined with the cost of actual tests, makes this option impractical. Herein, we compare some machine learning methodologies to propose a reliable strategy to detect people positive to COVID-19, analyzing saliva spectra obtained by Fourier transform infrared (FTIR) spectroscopy. After analyzing 1275 spectra, with 7 strategies commonly used in machine learning, we concluded that a multivariate linear regression model (MLMR) turns out to be the best option to identify possible infected persons. According to our results, the displacement observed in the region of the amide I of the spectrum, is fundamental and reliable to establish a border from the change in slope that causes this displacement that allows us to characterize the carriers of the virus. Being more agile and cheaper than reverse transcriptase polymerase chain reaction (RT-PCR), it could be reliably applied as a preliminary strategy to RT-PCR.