Professor Eduardo Eyras, along with his team and collaborators at the Australian National University’s John Curtin School of Medical Research has developed a tool to detect viral co-infections in COVID-19 patients to help develop informed strategies for antiviral treatment, vaccination and epidemiological control in the COVID-19 pandemic.
Various studies report that around 20 percent of SARS-CoV-2-positive individuals had a co-infection with other respiratory viruses and that these coinfections may increase the severity of the disease by altering disease progression and response to treatment.
Professor Eyras said identifying those co-infections will be relevant for treatment and prognostic purposes, however, the current standard of viral detection is based on PCR assays directed to SARS-CoV-2 only, which will miss possible co-infecting viruses. The new method, called PACIFIC, can identify SARS-CoV-2 and 361 other viruses at a sample concentration as low as 0.03%, with high specificity (low false-positive rate). “With PACIFIC, we are trying to provide a fast and easy-to-use, end-to-end tool that will enable researchers to monitor viral infections and co-infections to help manage the global COVID-19 pandemic.”
The new computational tool, based on artificial intelligence for natural language processing, detects viral co-infections from COVID-19 patient samples analysed with high-throughput sequencing (HTS). HTS enables the unbiased detection of RNA molecules present in a patient sample and can be performed on blood, sputum or any other patient sample.
Professor Eyras and his colleagues decided to develop a tool to identify SARS-CoV-2 and other potential viruses from the pool of HTS reads (i.e.: the fragments of the RNAs measured from the sample). To achieve this, the tool had to be able to identify the virus type from the RNA sequencing reads, and do this confidently enough for as many reads as possible.
“This is similar to language processing, where you recognise and classify texts, and organise them into organised groups,” Professor Eyras said. “We trained this model with examples of sequences that come from multiple viruses and from humans, and the model learns to separate them.”