Is It Possible to Predict the Next Pandemic?

It’s been two years since an epidemic of Zika began in Brazil, three since the largest Ebola outbreak in history erupted in West Africa, eight since a pandemic of H1N1 flu swept the world, and almost a hundred since a different H1N1 flu pandemic killed 50 million people worldwide. Those viruses were all known, but no one knew when or where they’d trigger epidemics. Other diseases, like SARS, MERS, and HIV, emerged out of the blue.

Sick of being perpetually caught off guard, some scientists want to fully catalogue all viral threats, and predict which are likely to cause tomorrow’s outbreaks. The PREDICT project has been doing that for 8 years; with $100 million in funding from the U.S. Agency for International Development, they’ve discovered nearly 1,000 new viruses. The Global Virome Project is even more ambitious. Proposed in 2016, and still existing in concept only, it aims to find and sequence almost all the viruses in birds and mammals that could potentially spill over into humans.

The GVP estimates that around half a million such viruses exist, and finding them would cost $3.4 billion. With that hefty price tag would come security. In lofty language, the project promises to switch the world “from responding to outbreaks to proactively preparing for them” and to “mark the beginning of the end of the Pandemic Era.”

There’s just one problem, say Jemma Geoghegan and Edward Holmes, two virologists based on Sydney. It won’t work.

In a new paper, Geoghegan and Holmes argue that these projects aren’t going to help preempt pandemics, for the simple reason that there are just too many viruses. About 4,400 have been identified; millions more have not, and only a tiny fraction of these could conceivably jump into humans. “The GVP will be great for understanding more about viruses and their evolution, but I don’t see how it’ll help us work out what’s going to infect us,” says Geoghegan. “We’re only just coming to terms with the vastness of the virosphere.”

There are ways of narrowing down the culprit list. Many teams have tried to map geographical hotspots from which diseases are most likely to emerge, pinpointing areas with tropical forests and lots of mammal species. Others have tried to find features in viruses that make it easier for them to spread between people. But having tried this approach themselves, Geoghegan and Holmes argue it’s not very useful.

Partly, that’s because the results of such studies are too broad to narrow down the list of suspicious viruses in a helpful way. Partly, it’s because such work is based on past epidemics—events that are relatively rare, and so difficult to draw reliable patterns from. For example, Saudi Arabia comes out as mostly cold in maps of disease hotspots, and yet it’s where MERS virus recently jumped into humans from an unlikely host: camels. “We’re trying to predict really, really rare events from not much information, which I think is going to fail,” Geoghegan says.

Ultimately, the odds that a given virus will cause an outbreak depend on the virus itself, the animals that host it, the people who stand to contract it, and the environment that all of them live in. “Within each of these categories, there are so many variables that could influence disease emergence,” says Jennifer Gardy, from the University of British Columbia. “It’s hard enough to model the effect of any one, and these factors likely interact in ways that we can’t possibly understand just by looking at each of them discretely.”

It’s even difficult to work out whether the viruses we already know about are going to cause outbreaks. Ebola and Zika, for example, were discovered in 1976 and 1947 respectively, but both managed to catch the world unawares this decade. “This is the easiest kind of prediction to make,” says Kristian Andersen, from the Scripps Research Institute, and we’re still about 10 to 20 years from doing it well. Next up in difficulty: predicting whether a virus like H7N9 bird flu, which can infect humans but isn’t known to cause major outbreaks, will eventually do so. Again, Andersen says that this isn’t feasible now, but should be with more research.

But predicting whether a newly discovered animal virus could jump into humans and cause a pandemic “is simply impossible,” he says. “What you’re trying to predict is likely something that happens maybe once out of tens of billions of encounters, with one virus out of millions of potential viruses. You will lose your fight against the numbers.” Even machine learning—using computers to divine patterns in data that humans might miss—won’t solve the problem because there isn’t enough good data for the computers to sift through.

Proponents of predictive initiatives say it’s too early to discount such approaches. If the same complaints had been raised in meteorology a century ago, “we wouldn’t have created the data that lets us forecast the weather, which we can do pretty well now,” says Jonna Mazet, the global director for PREDICT who also sits on the Global Virome Project steering committee.

“Can we predict pandemics? The answer right now is no. But just because something is hard to predict does not mean we cannot quantify its risk in a useful, actionable way—a logic that the insurance industry profits from,” adds Barbara Han, from the Cary Institute of Ecosystem Studies. No predictions are perfect, but at the very least, we can put boundaries on what is likely.

Resources aren’t infinite, though, and public health is an area that’s historically underfunded. Geoghagen argues that it would be best to channel efforts into approaches that would do the most good. For her, that involves looking at the “fault lines” where humans and animals meet—regions where people are more likely to be exposed to animal viruses because they are chopping down forests, or setting up dense animal markets, or hunting wild creatures for meat, or moving around a lot because of political instability.

Mazet agrees, and says that the Global Virome Project plans to look for viruses precisely at such fault lines. They want to, for example, search blood and meat samples of bushmeat, or the urine or saliva of rodents that share human homes. “It’s not aimed at detecting every virus out there,” she says, but she admits that the team hasn’t done the best job in explaining that to their fellow virologists.

But Geoghagen and Holmes argue that searching for these viruses in animals is still “a Sisyphean exercise.” You’d find too many, with no way of accurately assessing their risk of jumping into us. The project, they say, would be better off focusing on people—the workers in the bushmeat trade rather than the meat itself, for example. “Humans are the best sentinels: A virus discovered in humans very obviously can replicate in that host, which will not be the case for myriad viruses identified through biodiversity surveys of other [animals],” they say.

Andersen agrees. For the moment, preempting pandemics isn’t possible; what matters is catching them as early as possible. “Forget about detecting the virus before it jumps. Forget even about detecting the first patient,” he says. “Detect the first cluster of cases.” That’s possible if health workers routinely search for viruses in people who live at disease fault lines, and the advent of portable, pocket-size DNA sequencers could make such searches a reality.

These goals shouldn’t be seen in opposition, though. Kevin Olival from EcoHealth Alliance, who works with PREDICT, says that it would be impractical to study all fault-lines. “We need tools to help us narrow down and target our resources to the locations, host species, and viruses of greatest concern,” he says. Projects like PREDICT and the Global Virome Project may not act as crystal balls for future outbreaks, but they “help us prioritize on-the-ground disease surveillance.”

And PREDICT, through its work on detecting new animal viruses, has also helped to develop analytics tools and strengthen labs in developing countries, which will make it possible to do the kind of surveillance that Geoghagen, Andersen, and others are calling for. Everyone agrees that’s vital. “If we can’t even get routine surveillance working in hot-spot settings, we have no chance of getting something even more complex, like prediction, in place,” says Gardy.