21. Juni 2021, 10:00 Uhr | Tobias Schlichtmeier
Deep Learning detects viral infections and predicts acute – severe outbreaks.
AI methods such as deep learning are opening more and more new fields of application, for example in research. The University of Zurich, for example, has set itself the goal of detecting viral infections more quickly. Even serious infections should be able to be predicted in the future.
When viruses infect a cell, this leads to changes in the cell nucleus. These can be visualized using fluorescence microscopy. Researchers at the University of Zurich have trained an artificial neural network with such images in such a way that the algorithm reliably recognizes those cells that are infected by adenoviruses or herpes viruses. It also identifies acute, severe infections in advance.
In humans, adenoviruses can infect the cells of the respiratory system, while herpes viruses can infect those of the skin and nervous system. In most cases, this does not lead to the production of new virus particles, as the viruses are intercepted by the immune system. However, adenoviruses and herpesviruses can cause permanent, persistent infections that are only incompletely controlled by the immune system and produce viral particles for years. Similarly, these viruses can lead to sudden, violent infections. In this case, affected cells release large quantities of viruses – leading to infections that spread rapidly. The consequences are serious acute diseases of the lungs or nervous system.
The research group of Urs Greber, professor at the Institute of Molecular Biology at the University of Zurich (UZH), is the first to show that a machine-learning algorithm can identify those cells infected with herpes or adenoviruses based solely on the fluorescence of the cell nucleus. »Our method not only reliably identifies virus-infected cells, but also detects virulent infections in advance with high accuracy«, Greber says. The study authors are convinced that their development has many applications – such as predicting how human cells will react to other viruses or microorganisms. »The method opens up new ways to better understand infections and to discover new active agents against pathogens such as viruses or bacteria«, Greber adds.
The analysis method is based on the combination of fluorescence microscopy in living cells and Deep Learning. Herpes and adenoviruses that form inside an infected cell change the organization of the cell nucleus – these changes can be visualized with the microscope. To detect them by machine, the group uses a Deep Learning algorithm, more precisely a neural network. The researchers train it with a large set of microscopy images and extract patterns characteristic of infected or uninfected cells. »Once training and validation are complete, the neural network automatically detects virus-infected cells«, Greber said.
In addition, the scientists show that the algorithm can identify acute and severe infections with 95 percent accuracy and up to 24 hours in advance. Images of living cells from so-called lytic infections serve as training material. In this case, the virus particles multiply explosively and the cells dissolve. In addition, images of persistent infections, in which viruses are produced continuously but only in small quantities, serve as training material. Despite the great precision, it is still open which features of infected cell nuclei the artificial neural network recognizes to distinguish between the two phases of infection. However, it already allows to study the infection biology of infected cells in more detail.
The group has already discovered some differences: the internal pressure of the cell nucleus is greater during virulent infections than during persistent phases. In addition, a cell with lytic infection accumulates viral proteins more rapidly in the nucleus. »We therefore suspect that sophisticated cellular processes determine whether or not a cell disintegrates after viral infection. We can now investigate these and other questions«, says Greber.