New method accurately detects pathogenic infections in advance

When the virus infects cells, changes occur in the cell nucleus, which can be observed with a fluorescence microscope. Researchers at the University of Zurich have trained artificial neural networks to reliably recognize cells infected with adenovirus or herpesvirus using fluorescent images from living cells. This procedure also identifies severe acute infections in the early stages.

In humans, adenovirus can infect cells in the respiratory tract, and herpesviruses can infect cells in the skin and nervous system. In most cases, the virus is suppressed by the immune system, so it does not produce new virus particles. However, adenovirus and herpesviruses can fail to completely suppress the immune system and cause persistent infections that produce viral particles over the years. These same viruses can also cause sudden, violent infections in which affected cells release large amounts of the virus, resulting in a rapid spread of the infection. This can lead to serious acute illnesses of the lungs and nervous system.

Automatic detection of virus-infected cells

Urs Greber’s research group, a professor of molecular life sciences at the University of Zurich (UZH), found that machine learning algorithms used herpes or adenovirus-infected cells to fluoresce cells.

Our method not only reliably identifies cells infected with the virus, but also accurately detects pathogenic infections in advance. “

Professor of Molecular Life Sciences, University of Zurich, Urus Graber (UZH)

The authors of the study believe that their development has many uses, including predicting how human cells react to other viruses and microorganisms. “This method opens up new ways to better understand infections and discover new activators against pathogens such as viruses and bacteria,” Greber adds.

The analytical method is based on a combination of fluorescence microscopy of living cells and a deep learning process. Herpes and adenovirus formed in infected cells change the tissue of the nucleus, and these changes can be observed under a microscope. This group has developed a deep learning algorithm (artificial neural network) to automatically detect these changes. The network is trained with a large set of microscopic images that learn to identify patterns that are characteristic of infected or uninfected cells. “Once training and validation is complete, the neural network will automatically detect the cells infected with the virus,” Greber explains.

Reliably predict severe acute infections

The research team also demonstrated that the algorithm can identify acute and severe infections up to 24 hours before with 95% accuracy. Images of live cells with lytic infection, in which virus particles rapidly proliferate and cells are lysed, and images of persistent infection, in which only a small amount of virus is continuously produced, were used as training materials. Although this method is very accurate, it is not yet clear which features of the infected cell nucleus are recognized by the artificial neural network to distinguish between the two stages of infection. But without this knowledge, researchers have been able to study the biology of infected cells in more detail.

This group has already discovered some differences. Internal pressure in the nucleus is higher in toxic infections than in persistent stages. In addition, viral proteins accumulate more rapidly in the nucleus in lysed cells. “Different cell processes are likely to determine whether cells collapse after infection. We are now able to investigate these and other questions,” Greber said.