Back in 2016, four years before a pandemic brought the world to a standstill, the United Nations Environment Program (UNEP) sounded the alarm about zoonotic diseases, identifying them as Major Emerging Issues of Global Concern.
Now, according to World Health Organization, about 1 billion cases and millions of deaths each year are the result of zoonotic diseases, in which pathogens are transmitted from vertebrates to humans. Of the 30 new human viruses discovered in the past 30 years, 75 percent came from other animals.
But scientists at the Université de Montréal believe their new artificial intelligence model’s ability to highlight and predict emerging viral “hotspots” for observation could lead to advances in possible outbreaks of animal-to-human infections, and ideally With this, anything like COVID-19 can be prevented from happening again.
The algorithm, which took the researchers three years and 10,000 computing hours, was able to identify 80,000 new potential interactions between viruses and hosts, and where they are most interesting in the world.
“We have been working on this project since the first months of 2020, before the pandemic started,” said Timothée Poisot, a professor at the Université de Montréal’s Department of Biological Sciences.
Using machine learning, rather than manually linking data, the algorithm was able to evaluate thousands of mammalian species and thousands of viruses and calculate all possible combinations.
“The fundamental problem is that we only know one to two percent of the interactions between viruses and mammals,” Poisot said. “The network is fragmented, with few interactions, concentrated in a few species. We want to know which viruses are likely to infect which mammals so we can determine which interactions are most likely to occur.”
The team used CLOVER, the largest open dataset, which describes 5,494 interactions between 829 viruses and 1,081 mammalian hosts, most of which focused on wild animals, and several other datasets, including Host-Pathogen Phylogenetic Project (HP3), Enhanced Infectious Disease Database (EID2), and Global Mammalian Parasite Database V2.0 (GHMPD2).
“Some of the datasets we had were older: they contained outdated names for certain species, or there were errors because the data was entered by hand,” Poisot said of the time-consuming process required for the process. machine learning. “After that, the main task is to determine our level of confidence in the model’s ability to make predictions.”
The researchers then focused on 20 viruses that were deemed of concern and could potentially infect humans.
“We had a lot of discussions in the group because some of the results seemed strange to us at first,” said Poisot, who was surprised to see the mouse-related results. outer segment virus identified as worthy of attention. “We were skeptical, but when we searched the literature, we found that there had been cases in humans.”
The researchers were also able to pinpoint regions through the model, which could help scientists conduct more targeted virus and vaccine research.
“Our model makes spatial predictions, but more precisely, the model specifies in which groups of mammals and where certain types of viruses are likely to be found,” Poisot said.
The results revealed two regions of particular interest: the Amazon, where virus and host interactions are more primitive and where new interactions are most likely to be seen; and sub-Saharan Africa, where the algorithm identified potential carriers of zoonotic new host for the virus.
“We’re really changing where we need to go to study mammals to discover new viruses,” Poisot explained.
While zoonotic pathogens can take many forms—bacteria, parasites, viruses—their prevalence is expected to become increasingly common as humans and nonhuman animals continue to occupy more of the same space.
The team hopes its model will not only provide a new starting point for research, but real-world monitoring as well. The next step is to take this artificial intelligence to the next level, including more microbiological, immunological and ecological mechanisms, for a more comprehensive understanding of the global virome.
“The algorithm takes our known network and projects it into a new space, sort of like a shadow puppet show: It reveals interactions in a new way,” Poisot said. What kind of virus. “
The study was published in the journal pattern.
source: University of Montreal