Five years after starting out as an experimental project to see if advanced sensors and artificial intelligence could spot the signs of a disease outbreak before it happens, Microsoft Premonition is turning into an honest-to-goodness biothreat protection network.
Premonition’s researchers aim to set up about 100 sensor stations in Texas’ Harris County, to track swarms of mosquitoes that could transmit diseases ranging from malaria and dengue fever to Zika and West Nile viruses. AI algorithms will analyze that tracking data for the telltale signs of an epidemic in the making, just as weather forecasting programs look for the signs of a storm on the way.
“It will really be almost like a weather map, the likes of which has not really been seen before in the mosquito vector space,” Ethan Jackson, senior director of Microsoft Premonition, told Geekwire.
The expansion of the Premonition program was announced today in conjunction with this week’s annual Microsoft Ignite conference for software developers.
Harris County, dominated by the city of Houston with a population of 4.7 million, was the site for Microsoft Premonition’s earliest experiments. Those field tests were aimed at determining whether sensor-equipped mosquito traps could identify the particular types of mosquito that causes particular diseases — for example, Anopheles mosquitoes for malaria, Aedes mosquitoes for Zika and dengue, Culex mosquitoes for West Nile.
Microsoft researchers found that they could, based on data from optical sensors that tracked the beat of the mosquitoes’ wings. During 2016’s field tests in Harris County, the system recorded a 90% accuracy rate for identifying the mosquitoes linked to the Zika virus. That’s no small feat, considering that there are 3,600 known species of mosquitoes, 50 of which are active in Harris County.
Since then, Jackson and his colleagues have widened the system’s capabilities by bringing wild mosquitoes to a custom-built facility on Microsoft’s campus in Redmond, Wash., known as the Premonition Proving Ground.
“This is Microsoft’s first biocontainment facility,” Jackson said. “It’s unique in its ability to allow us to import wild species, rear them up from eggs, and then digitize them in our sensors so that we can pre-train our classifiers.”
Researchers can then check test their robotic traps and digitized models with real-life mosquitoes. “We have a feedback loop, an agile engineering loop that’s happening at Microsoft,” Jackson said.
The lab experiments helped Microsoft fine-tune Premonition’s tools.
“We’re at a place now where we’re ready to go back and try these systems at scale, and see what happens when you instrument this city with these kinds of systems so you’re getting 24/7 monitoring of the biome,” Jackson said.
Umair Shah, executive director of Harris County Public Health, hopes the technology’s impact won’t be limited to mosquito-borne diseases.
“This partnership will also evaluate new genomic capabilities to detect known and emerging pathogens from environmental samples, which we now know is especially important for diseases like COVID-19,” he said in a Microsoft blog posting..
Being able to predict where an outbreak might blossom would help public health officials stay a step ahead — for example, by using pesticides in targeted areas to tamp down mosquito swarms, or ramping up localized strategies to catch coronavirus before it spreads.
One of Microsoft’s corporate partners on Premonition is Bayer, which is participating in an industry campaign to eradicate malaria by 2040. Jacqueline Applegate, president of Global Vegetable Seeds and Environmental Science at Bayer, said Microsoft Premonition will help Bayer “be even more prescriptive and optimize our vector control strategies so that they have the greatest impact.”
Over the past five years, Microsoft Premonition’s technologies have been tested at sites ranging from the sands of the Florida Keys to the forests of Tanzania. Jackson said the follow-up experiments have shown that sensor networks can pick up a wide spectrum of information about biological interactions in ecosystems.
To explore the wider applications, Microsoft is partnering with academic researchers at Vanderbilt University, Johns Hopkins University, the University of Pittsburgh and the University of Washington’s Institute for Health Metrics and Evaluation. This month, the academic consortium began a $924,000 project funded by the National Science Foundation to develop predictive models of ecosystem-wide activity.
Jackson marveled at how his perspective on Premonition has progressed.
“When we first started this project, we were asking, how could you understand biological threats from the perspective of a mosquito? That was the science question that drove our initial engineering,” he said. “We’ve really evolved from that question to say there’s a set of technologies that form a sensor network, and that sensor network is the thing that’s missing today.”
After five years, Microsoft says it’s ready to fill in that missing piece.
