ABSTRACT
The most common COVID-19 testing relies on the use of nasopharyngeal swabs. However, this sampling step is very uncomfortable and is one of the biggest challenges regarding population testing. In the present study, the use of saliva as an alternative sample for COVID-19 diagnosis was investigated. Therefore, high-resolution mass spectrometry analysis and chemometric approaches were applied to salivary lipid extracts. Two data organizations were used: classical MS data and pseudo-MS image datasets. The latter transformed MS data into pseudo-images, simplifying data interpretation. Classification models achieved high accuracy, with pseudo-MS image data performing exceptionally well. PLS-DA with OPSDA successfully separated COVID-19 and healthy groups, serving as a potential diagnostic tool. The most important lipids for COVID-19 classification were elucidated and include sphingolipids, ceramides, phospholipids, and glycerolipids. These lipids play a crucial role in viral replication and the inflammatory response. While pseudo-MS image data excelled in classification, it lacked the ability to annotate important variables, which was performed using classical MS data. These findings have the potential to improve clinical diagnosis using rapid, non-invasive testing methods and accurate high-volume results.
