Boffins have demonstrated that machine-learning algorithms may be able to help scientists identify viruses, and could even be used to develop more efficient tests for the presence of the COVID-19 coronavirus in the near future.
Researchers at Nagoya University, Osaka University, and the National Institute of Advanced Industrial Science and Technology, all in Japan, trained a system to recognize five different types of viruses: respiratory syncytial virus, coronavirus, adenovirus, and influenza A and B. These bio-nasties attack respiratory systems, with typical cold and flu symptoms.
Their paper, published in ACS Sensors, describing the AI system claims the algorithms used are capable of identifying these virus types with 99 per cent accuracy. First, the team built a “nanopore sensor,” a device fashioned from a silicon wafer. Tiny holes 300nm across are drilled into the metal to form channels that allow ions to travel through the wafer when a voltage is applied.
The number and rate of ions passing through the device can be detected by the strength of the current generated. Virus particles block the holes, making it more difficult for the ions to move, and the amount of current produced declines. The dip provides a nifty way of detecting viruses, since it provides hints describing the particle’s properties. In fact, the characteristics of the dip reveal the type of virus present, it is claimed.
“The nanopore was designed to make the electricity sensitive to multiple physical properties of the viruses such as the size, shape, surface charge density, and mass,” Makusu Tsutsui, co-author of the paper and a professor at Osaka University, explained to The Register. “All of these features were reflected in the fine profiles of the resistive pulse signals measured.”
Just cough into your phone, please… MIT lab thinks it can diagnose COVID-19 from the way you expectorate
In other words, the amount of current that decreases in the nanopore device is unique for each type of virus. The current profiles provide a way to characterize the viruses, and rotation forest algorithms in the machine-learning system were trained to classify each type of virus from these profiles.
The sensor can detect novel viruses by checking if the current profile matches up with any known samples. “The nanopore approach may become applicable to identify new strains. Although this of course depends on how different they are from the existing strains, we already found in our previous study that the method is sensitive enough to discriminate influenza allotypes,” Tsutsui said.
The team reckons their method is particularly suited for identifying coronaviruses since the shape of each strain differs slightly. By training the nanopore sensors to recognize the COVID-19 virus, they believe the device can be developed further to test for the presence of the novel coronavirus in samples. “This work will help with the development of a virus test kit that outperforms conventional viral inspection methods,” said last author Tomoji Kawai earlier this week.
“Compared with other rapid viral tests like polymerase chain reaction or antibody-based screens, the new method is much faster and does not require costly reagents, which may lead to improved diagnostic tests for emerging viral particles that cause infectious diseases such as COVID-19,” Tsutsui added. ®