AI used to 'predict the next coronavirus'

 

British researchers have developed an algorithm that has identified mammals most likely to be coronavirus reservoirs from which the next viral threat could emerge. Their findings greatly increase the number of species to watch, including domestic cats.

There are far more coronavirus hosts in the animal world than previously thought. In any case, this is the conclusion reached by artificial intelligence (AI) that has been commissioned by a team of scientists from the University of Liverpool to go after the mammals likely to give birth to the future Sars, Sars-CoV-2 or Mers-CoV.

Race result: "There are 40 times more species likely to have at least four forms of coronavirus in them than was already established, and about 30 times more mammals in mammals that could recombination of Sars-CoV-2 [to give rise to a new virus]," the researchers wrote in a paper published Tuesday, February 16, in the journal Nature Communication.

All mammals under the magnifying glass of AI

They developed this algorithm to fill in the gaps "in our limited knowledge of which mammals are potential hosts for which coronaviruses, which is necessary to understand where recombinations of these viruses may occur to form new ones," explains Marcus Blagrove, a virologist at the University of Liverpool and one of the co-authors of the article, contacted by France 24.  The proof: "Before the arrival of Sars-CoV-2, there was very little interest in pangolin," adds Maya Wardeh, lead author of the study, also contacted by France 24.

There are simply too many mammals on Earth for humans to study them all. Not to mention that "some are in countries that do not have the budgets to do this kind of research," Wardeh said.

An obstacle that is not likely to frighten an algorithm fed to "big data". The two researchers provided their home AI with a considerable amount of data on all listed mammals - biological proximity to other animals known to be coronavirus carriers, the geographical area where they live, or life expectancy - and did the same with known viruses.

The machine then provided the most complete map to date of all rodents, carnivores, bats, or even artiodactyls (cattle, hippos) that can be reservoirs to several coronaviruses. Specifically, "the model calculates the probability for each mammal species to be infected with each of these viruses, which then allows us to predict which ones may be hosts for multiple coronaviruses at once," Wardeh explains.

On average, each coronavirus can have 12.5 carrier mammals and each animal studied can be infected with 5.5 different viruses, based on these calculations.

An effective algorithm

For Sars-CoV-2, the AI has estimated that there are 126 potential hosts in the wild, i.e.... 122 more than was accepted when the Liverpool researchers got to work in the summer of 2020. For example, the small Asian yellow bat, hedgehog or garden rabbit should be added to the list of potential carriers of Sars-CoV-2. 

This work has also identified the most important coronavirus "marmites" in the animal kingdom. Bats, of course, are at the top of the list. Some of them may be infected with 68 different coronaviruses. This is the case of the great rhinology, or large horseshoe, the largest species of bat found in Europe.

More surprisingly, the domestic cat has also been identified as particularly permeable to these viruses by the British researchers' algorithm. It is a potential host for 65 coronaviruses, including Sars-CoV-2. This feline is, for the authors of the article, one of the main "underestimated risks" to become the cradle of a new virus, the result of a mixture of those that would be present in its body. 

Certainly, these results are the result of machine calculations and no field observations. "Obviously these findings are not 100% accurate," admits Maya Wardeh. But both researchers are confident in the robustness of their model. Since they developed it, the scientific community has discovered several new hosts of Sars-CoV-2, such as mink, and "in most cases, they were identified by the algorithm," Blagrove said.

That doesn't mean you have to stop petting your cat!" says Maya Wardeh. Just because the kitty is potentially very permeable to a large number of coronaviruses does not necessarily mean that it will catch them. In particular, it should be in the vicinity of another mammal that transmits the virus to it, and for a pet, which often leads a casanière life, such opportunities are rare.

Besides, this model does not predict the likelihood of new coronaviruses appearing in these leg tanks. "All we do is indicate in which animals this can happen. Calculating the actual risk is infinitely more complex," Blagrove says.

Their work also does not indicate whether the result of possible recombination will be a more virulent Sars-CoV-2 or one of the many coronaviruses that, at worst, causes a cold in humans. "It would take a bioinformatician capable of anticipating the characteristics of the result of this possible mixture," explains the virologist.

The algorithm's predictions can give the impression of an animal kingdom even more dangerous than it was for our health. But that is not the aim of the efforts of British researchers. "We are just describing a phenomenon that already exists in nature - the recombination of coronaviruses - and specifies which species we think should be monitored in particular," Blagrove says. And so not to be, as with sars-CoV-2, caught off guard that the coronavirus came.