Reservoir rats

BETWEEN 1346 and 1353 the Black Death killed over a third of Europe’s population. It took 150 years for the continent to recover. The disease was so devastating that it changed the social order, as a scarcity of labour led to higher wages for the survivors, hastening the demise of feudalism.

The plague was caused by Yersinia pestis, a bacterium that lives in fleas. In Europe, those fleas lived mostly on black rats (pictured). In Asia, where the disease came from, they lived on gerbils. It was thus a zoonotic illness: one usually carried by animals, but which infects people when given the chance. Since human beings have little evolutionary experience with such illnesses, and therefore little resistance to them, they can be particularly dangerous. Ebola fever is a zoonosis. So, as their names suggest, are the swine- and bird-flu strains that keep epidemiologists awake at night.

Trying to work out which animals are reservoirs of disease that might infect humans is therefore an important job. It is also a tricky one. There are lots of animal species, a lot of unpleasant viruses and bacteria, and not enough zoologists and doctors to sort through them all. But Barbara Han of the Cary Institute of Ecosystem Studies, in New York, and her colleagues think they have a way to help with this labour shortage. They propose to apply artificial intelligence (AI) to the problem.

As they describe in a paper in the Proceedings of the National Academy of Sciences, they have used a form of AI called machine learning in the search for reservoir species. Machine learning lets computers study large sets of data and identify patterns and organising principles. It is a hot topic among the world’s big technology firms, such as Google and Amazon, who have plenty of data to make sense of. Dr Han and her colleagues set their computers the task of looking at rodents, and searching for rules that describe which ones are likely to harbour and transmit potentially zoonotic diseases.

Zoologists recognise more than 2,200 species of rodent. Of these, 217 are known reservoirs of pathogens (ranging from viruses and bacteria to protozoa and worms) that have an appetite for humans. At least 79 carry more than one such pathogen.

The team used 86 criteria to characterise rodent species, including how fast they breed, the types of habitat they live in and the specifics of their physiology. (Those data, she points out, were painstakingly collected, by hand, by generations of zoologists on field trips—a retort to the question of why scientists should spend taxpayers’ money investigating the reproductive history of some obscure animal or other.) They also classified rodent-borne bugs in a similar way. They then fed this smorgasbord of data, far more than any human being could hope to process, into their computers, and told the machines to look for combinations of traits that predict whether a particular species is likely to harbour something that can cause human disease.

They duly found some. Reservoir species, the machines said, tend to have speedy lives, short gestation periods, big litters and early sexual maturity. They also have large ranges, inhabit areas with few other rodent species as neighbours, and (not surprisingly) live near big human populations. Once trained, the computers could predict known reservoir rodents with better than 90% accuracy.

The researchers’ next step was to use the model the computers had come up with to make a list of species not known as reservoirs, but which might be. When they did this, they found 150 that fitted the bill. Top of the list in North America, for example, were the montane vole and the northern grasshopper mouse. The model also suggested that about 50 species already known to spread at least one human disease were likely to spread others, too.

These conclusions are statistical rather than definitive. They cannot say for sure that any particular species is a reservoir of diseases which could be transmitted to people. But they do imply that these species would be worth looking at. And Dr Han says that the computers’ conclusions fit with fieldwork and evolutionary theory.

Species that live fast and breed rapidly, she says, devote fewer resources to fighting diseases than their more sedate relatives. Their evolutionary strategy is to deal with illnesses by outbreeding them, accepting that a large proportion of individuals will be infected at any given time. This boosts her confidence that the computer models are indeed onto something.

She and her colleagues were also able to pinpoint areas at particular risk of harbouring reservoir rodents. Large swathes of the world probably have at least one such species. But the researchers found hotspots, too, where as many as eight or nine potential threats may be lurking. One was Central Asia, in modern-day Kazakhstan and northern China, which has already been fingered by historians as a likely source of the Black Death.

The other, though, was a surprise. It was the American states of Kansas and Nebraska. Zoologists looking to make a name for themselves—and to help stave off future epidemics—may therefore want to make the prairies the destination of their own obscure-rodent-studying field trips.