The corona virus caused the worst global epidemic in the last hundred years. Earlier, smaller outbreaks include avian flu, SARS, Ebola, and the monkeypox disease that made headlines recently, although it has not spread widely in developed countries.
With the intensifying climate changes, the increase in the global population, the reduction of the habitats of wild animals and their pushing to the proximity of human population centers, scientists have long predicted the appearance of an epidemic like the corona, and even predicted that there will be more like it. In the coming decades, epidemics resulting from the transfer of new viruses from animals to humans are expected more frequently than before. Some of them will be more limited or less violent than the corona, but some may be worse than it, both in the degree of infection and in the degree of lethality.
Scientists are trying to build models that will warn of possible virus outbreaks, and recently enlisted the help of artificial intelligence. This artificial intelligence should search for and quickly identify critical threats, but further development is required for the technology to mature.
The bird flu that appeared in 1997 demonstrated the transfer of viruses from animals to humans. Chickens in the coop Philippe Benoist, Eurelios, Science Photo Library
Foundation of Efforts: Encyclopedia of Viruses
If we get to know the variety of viruses that exist in animals better, study them and compile a kind of “encyclopedia of viruses”, which centralizes and makes the information about them accessible, we will be able to know which viruses we might be dealing with, even before they break out as epidemics that affect humans. For example, if we identify a place in the world with a virus that could cause an epidemic, we can apply restrictions and control to the area in advance. We can limit the contact between humans and wild animals at the risk points, or establish sanitation conditions for these areas and monitor compliance with them, thus preventing the outbreak. Scientists will be able to begin preliminary research on suspected viruses in order to speed up the development of vaccines, drugs and treatments that support in the event of an epidemic, thus saving many lives.
For this purpose, the international PREDICT project was launched in 2009, which was financed with 200 million dollars by the American Agency for International Development (USAID). The project continues until 2020, and the researchers identified 949 new viruses that infected humans, farm animals and wild animals in 34 countries.
One of the studies in the project, published in 2017, predicted the corona epidemic: the study estimated that there are thousands of undiscovered corona viruses in the bodies of bats, and predicted that the Southeast Asian region would be the most diverse habitat for viruses from the family that includes the SARS-CoV-2 virus, which is going to cause to the corona epidemic. The study identified that in places where there is intense contact between humans and wild animals, corona viruses are common in bats, and it seems that the origin of the corona epidemic is in a market in such an area.
Another study examined the compatibility between virus strains and the animals they infect. Ecologist Kevin Olival from the EcoHealth Alliance in New York, who led the study, told Nature: “The goal was to understand which viruses are able to infect people, from which animals we get the most new viruses and what are the factors underlying these patterns.” The study found a connection between the frequency of human infection and the biological proximity to infectious animals and between factors that affect the level of contact between humans and wild animals in those areas, such as the density of the human population in the wild animals’ habitat.
The research team used statistical models to predict which animals are likely to be carriers of zoonotic viruses, those that may pass from animals to humans, and in which areas this is likely to happen. Significant animals in this context were bats, rodents and primates – great apes such as orangutans and chimpanzees – and the significant regions included South America, Africa and Southeast Asia. The researchers also identified some of the characteristics of a zoonotic virus, for example the range of species it can infect.
One of the studies in the project predicted the corona epidemic in 2017. Corona virus Image: Kateryna Kon, Science Photo Library
significant limitations
For all its achievements, PREDICT was a running program only. The scope of the viruses it covered was “a drop in the ocean,” according to Olivelle. Scientists proposed in 2016 to establish a global virus project that would continue the research in an international partnership of governments and non-governmental bodies. The follow-up project was supposed to map most of the virus strains present in mammals and birds, since these are the departments from which most viruses come to humans, but the project did not receive funding and was not implemented.
Researchers who opposed the project argued that it was on an impossible scale. It is not known how many virus species currently exist in mammals and birds and there is no consensus regarding the estimates. Some scientists estimate that the number is about 1.67 million, of which at least 320 thousand are in mammals alone. By 2020, only about 4,000 viruses have been identified, and they change rapidly, so a one-time research effort will not be enough and it will be necessary to invest great efforts on a regular basis to update the database.
It has also been argued against the project that even if the genetic sequence of a virus is deciphered, there are unknown factors that may affect its likelihood of causing an epidemic, such as its infectivity factor in humans. Also, there is the sampling bias of the studies: there is time and resources to investigate in depth only a tiny proportion of the vast variety of existing viruses, and the tendency is to study families of viruses that have already spilled over to humans. This makes sense when looking for a cure or a vaccine for a virus that has spread, but it does not contribute to predicting the next epidemic and it may even divert attention from other, no less reasonable research directions.
The research uses statistical models to predict which animals are likely to be carriers of viruses that can be transmitted to humans, and where. Bat meat for sale in the market in Indonesia Sony Herdiana, Shutterstock
The AI ​​prioritizes targets
This is where artificial intelligence may contribute. Researching a single virus can take a long time, while artificial intelligence may identify and mark high-priority targets for in-depth research, when it has little and easy-to-obtain information. Often, the first thing discovered about a new virus is its genetic sequence, and obtaining this information requires hours to a few days.
Computational virologist Nardus Mollentze from the University of Glasgow developed a model that uses as a benchmark the genetic similarity of the virus to parts of human DNA. The logic behind this benchmark is that viruses that, during evolution, develop genetic segments similar to those of the host, reproduce better or hide from the immune system better. The model was tested on about 900 viruses and was able to identify with 70 percent accuracy which of them are zoonotic.
Molenza continued the research with researchers from the Verena Institute in the United States, integrated into the model the ability of viruses to learn to live in surrogates and improved the accuracy of the model to about 80 percent. The hope is that in the future it will be possible to gather knowledge about the interaction between the virus and the host at the molecular level.
Not all models are equally effective. There are models that have learned to classify cases only according to patterns, and will have difficulty predicting new cases. In contrast, there are models that are able to deduce the reasons for these patterns and will be more successful in forecasting. Differentiating between the types of models is difficult, and the ability to do so is expected to be essential. Colin Carlson, a biologist at Georgetown University in Washington who directs the Verna Institute, put it this way in an interview with Nature: “This is the question: are we just teaching the machines to repeat what they know, or are they learning principles that they can take Strong for a new space?”
Molenza’s model uses the genetic similarity of the virus to parts of human DNA. A bat was tested for the presence of the Ebola virus Philippe Psaila, Science Photo Library
The challenge: collecting cross-sectional information
To perform well, AI models need a database that is both large and high quality. A model required to predict whether a particular virus is zoonotic needs to be trained on many and varied examples. They should include viruses that have been transmitted to humans and viruses that have not been transmitted. They should include adequate representation that will eliminate bias from external factors such as the degree of exposure of humans to the virus. In virology, the information is few and lacking compared to the variety that exists in reality. Only a few leaks happen each year, so this makes it difficult for the algorithms to effectively learn and predict the future. Scientists decide which viruses to collect information on, based on their knowledge of the highest risks.
In order to train the models optimally, information must be collected on the geographical distribution of the viruses and their taxonomy – the division into groups according to common characteristics, which includes division into classes, families and species. For this purpose, we should not focus on viruses that we estimate to be dangerous, but on regions and families about which we know the least.
To get closer to achieving the goal, cooperation between scientists and research institutions all over the world, free flow of data and uniform and agreed protocols for data collection are needed. Here the obstacles are more political and cultural than scientific, for example mistrust between countries, conflicting public priorities, and academic incentives that favor publication over large-scale data collection. The solutions to overcome these obstacles are also more political and cultural than scientific. About this, Olivell told Nature: “This is the main issue and to deal with it, building trust is required. We need to make sure that we give value, not only in vaccines but also in training, capacity building and partnership in articles.”
In order for artificial intelligence to be able to give accurate predictions, close cooperation and mutual trust between scientists, research institutions and governments is necessary. Illustration of people, relationships and points of contact between them Immersion Imagery, Shutterstock
Artificial intelligence will filter out viruses
The ambition is that the computer models will provide available tools that will allow the decision-makers and the general public to understand in advance what the significant threats from zoonotic viruses are and which epidemics may break out. At a more advanced stage, we may be able to use artificial intelligence to classify the possible epidemics according to their degree of severity and invest resources and efforts accordingly. The corona virus taught us how important the period of time at the beginning of the epidemic is in order to contain it.
Much more research is required to build the information bases of the viruses and to understand the characteristics that allow an artificial intelligence model to develop a high predictive ability. However, even when the technology matures, it probably won’t accurately predict every outbreak. The hope is that we can predict epidemics the way we predict the weather: narrowing down the space of possibilities from a vast and unknown collection to a small number of realistic scenarios and better gauge the likelihood of each scenario.