Noninvasive and Reagent-Free Technique Uses Raman Spectroscopy and Machine Learning for Detection of COVID-19

By LabMedica International staff writers
Posted on 10 Feb 2022

Image: Machine-learning model (Photo courtesy of Ember et al., doi 10.1117/1.JBO.27.2.025002)

Researchers have developed a new and improved method that uses Raman spectroscopy and machine learning for the detection of SARS-CoV-2.

The noninvasive and reagent-free technique for the efficient detection of COVID-19 has been developed by biomedical researchers at Polytechnique Montréal (Montreal, Canada). Reverse transcription polymerase chain reaction (RT-PCR) techniques are currently the gold standard for detecting SARS-COV-2, the virus that causes COVID-19, although they have certain limitations. RT-PCR involves the transportation of samples to a clinical laboratory for testing, which poses logistical difficulties. It also requires the use of reagents, which could be in short supply and may be less effective when the virus mutates. Moreover, RT-PCR tests can be time-consuming and less sensitive in asymptomatic individuals, rendering them unfeasible for widespread rapid screening. Hence, researchers are trying to devise novel methods for better detection of COVID-19 infections in point-of-care settings, without the need to send away samples for testing.

The new reagent-free detection technique that is based on machine learning and laser-based Raman spectroscopy uses saliva samples. Unlike nasopharyngeal swabs, saliva sampling is safer and noninvasive. Raman spectroscopy is routinely used by researchers to determine the molecular composition of samples. Put simply, molecules scatter incident photons (particles of light) in a unique manner that is dependent on underlying chemical structures and bonding. Researchers can sense and identify molecules based on their characteristic Raman "fingerprint" or spectrum, which is obtained by shining light at samples and measuring the scattered light.

COVID-19 can cause chemical changes in the composition of saliva. Based on this knowledge, the research team analyzed 33 COVID-19-positive samples clinically matched with a subset of a total 513 COVID-19-negative saliva samples. The Raman spectra they obtained were then trained on multiple-instance learning models, instead of conventional ones. The results from this method indicate an accuracy of about 80%, and the researchers found that taking sex at birth into consideration was important in achieving this accuracy. Although saliva composition is affected by time of day as well as the age of the test subject and other underlying health conditions, this technique can still prove to be a great candidate for real-world COVID-19 detection. These findings can facilitate better COVID-19 detection in addition to paving the way for new tools for other infectious diseases.

"Our label-free approach overcomes many limitations of RT-PCR testing. We are working to commercialize this as a faster, robust, and low-cost system, with potentially higher accuracy," said Katherine Ember, a postdoctoral researcher at Polytechnique Montréal, Canada, and first author of the study. "This could be easily integrated with current viral detection workflows, adapted to new viruses and bacterial infections, as well as accounting for confounding variables through new machine learning approaches. In parallel, we are working on reducing the testing time further by using nanostructured metallic surfaces for containing the saliva sample."

Related Links:
Polytechnique Montréal