Machine Learning: A New Tool for Predicting Reservoirs of Zoonotic Diseases

Predicting an outbreak of an infectious disease before it occurs is no longer science fiction—or almost not. In a recent study*, researchers used a computer capable of modeling large datasets to predict which animals will carry potentially zoonotic viruses, bacteria, or fungi. These predictions should help improve disease prevention and response to outbreaks.   Most new outbreaks of infectious diseases occur when a virus, bacteria, or fungus jumps from animals to humans. Accurately predicting when and where these zoonoses will occur could prevent them from becoming epidemics. However, maintaining active surveillance of diseases worldwide is costly and time-consuming. To support research in this field, a team of scientists has developed a machine learning program capable of analyzing a database containing information on hundreds of mammal species, including the geographic range of their habitats and their reproductive strategies. Their program evaluated some 86 different predictive variables—such as body size, lifespan, and population density—to uncover common patterns among animals identified as reservoirs or vectors of zoonotic diseases. To simplify the results, the team limited its analyses to rodents, which carry a high number of zoonotic agents, ranging from Yersinia pestis to the rabies virus and hantaviruses. With over 90% accuracy, the machine learning program identified which rodent species are—and could become—new zoonotic reservoirs, as well as the geographic regions where emerging pathogens are likely to occur. The program also describes the biological profiles that characterize reservoir species. Thus, reservoir rodents exhibit a rapid reproductive strategy, with early sexual maturity, a short gestation period, and large litters. According to the authors, this short life cycle allows these species to successfully pass on their genes across vast geographic areas before the disease they carry proves fatal to them. Consistent with the modeled profiles, the program has so far identified more than 50 rodent species likely to be new reservoirs of zoonoses and predicted some 150 new infectious agents in species already known to be reservoirs of a single zoonotic agent. The geographic areas most likely to harbor these new reservoirs are the Midwest region of the United States (Kansas and Nebraska), the Middle East, and Central Asia (Kazakhstan and northern China). Ultimately, the predictions generated by these models provide a basis for directing disease surveillance in the field toward specific regions and species, as well as for testing new hypotheses about the reservoir role of wildlife. This study shows that it is now possible to accurately predict which wild species will carry zoonotic infections. As for preventing them, that is another story—one yet to be written by public health agencies and epidemiologists, who will have to rise to the challenge of controlling and treating previously unknown zoonoses.   * Barbara A. Han: Rodent reservoirs of future zoonotic diseases, PNAS, 2015, http://www.pnas.org/content/early/2015/05/14/1501598112.full.pdf