- A recent study demonstrates that machine learning methods could determine the risk of a viral jump or “spillover” from animals to humans using viral genomes.
- The research models predicted that genetically similar nonhuman primate viruses had an increased risk of human transmission, which was not the case with other animal groups.
- Scientists will need to conduct more research to confirm that the viruses that prediction models identified represent a high risk of animal-to-human transmission.
Zoonotic diseases, or zoonoses, occur due to viruses, bacteria, parasites, or fungi that spread between animals and people.
Approximately
Dr. Barbara A. Han, Ph.D., a disease ecologist for the Cary Institute of Ecosystem Studies, explained in a podcast, a “zoonotic disease is just an infection that originates in an animal […], caused by a parasite or pathogen that’s perfectly happy to live in this wild species.”
Dr. Han elaborated: “Occasionally, that pathogen or parasite will spill over into a human, and 99% of the time, that’s where it ends — that person might get sick, but it’s a dead-end host, so it doesn’t go any further. Some of them can transmit from person to another person, so that secondary transmission is really critical for something that has the potential to become pandemic.”
Therefore, improved global communication, coordination, and collaboration between human, animal, and environmental experts to prevent, detect, investigate, prioritize, and respond to zoonotic diseases is imperative.
This strengthened communication is vital to allow us to create an early warning system to prevent or mitigate the next pandemic.
This need led researchers from the University of Glasgow in the United Kingdom to formulate a new approach. They used viral and human
Their latest study appears in the journal PLOS BIOLOGY.
Approximately 1.67 million undescribed animal viruses cause infections in mammals and birds, and scientists believe that up to half could spill over into humans.
Dr. Nardus Molentze, a co-author of the study and research associate at the University of Glasgow’s Centre for Virus Research, spoke with Medical News Today:
“In recent years, the field of virus discovery has made significant advances, to the point where viruses previously unknown to science are being reported regularly. But this leads to a challenge — we still have a huge task ahead of us in terms of characterizing viral diversity in nature, and beyond discovery, to work out whether these viruses pose a threat.”
He added, “In 2018, my co-authors showed that RNA virus genomes contain enough signal for machine learning methods to identify the broad reservoir group — for instance bats, rodents, and primates — in which they circulate naturally.”
In other words, they showed that by analyzing a viral genome alone, their model could identify what type of animal with which the virus could cause infection.
Dr. Molentze continued, “This made us wonder whether […] virus genomes might also contain clues about their ability to [cause infections in] humans specifically when given the opportunity.”
The researchers collected a genome sequence from 861 RNA and DNA virus species from 36 viral families that can infect animals.
To investigate, they classified each virus according to its ability to cause infection in humans using information from three published datasets.
They also noted each virus’ similarity to viruses that can cause infections in humans and built machine learning models to predict whether these infections could occur.
The scientists tested several learning-based models to identify the best-performing model and used this to rank 758 virus species.
The machine-learning model correctly identified 70.8% of human viruses with high or very high zoonotic potential.
In a study of 645 animal viruses that were not part of the training data, the models predicted increased zoonotic transmission risk of genetically similar, or
A second experiment predicted the zoonotic potential of all currently recognized coronavirus species and the human and animal genomes of all severe acute respiratory syndrome-related coronavirus.
Dr. Molentze commented on the findings, “Our work shows a path through which virus discoveries can be turned into actionable information: the ability to identify which newly discovered viruses are most likely to be able to [cause infection in] humans with reasonable accuracy allows us to focus further characterization efforts on those viruses.”
Since the computer learning method requires only a genome sequence, it may provide a low cost approach for evidence-based virus surveillance.
Dr. Molentze added, “Our model is far from perfect — predictions will contain both false positives and false negatives, which can only be distinguished through further characterization of these viruses in the lab.”
Dr. Molentze emphasized the need for further study, “If we want to turn virus discoveries into actual pandemic preparedness, we need to characterize newly discovered viruses. […] Models […] could help prioritize viruses at various stages in this characterization pipeline, making their implementation more efficient and feasible, particularly if we are also able to develop methods predicting other aspects of risk, such as virulence and ability to transmit.”