Talk about building a better proverbial mousetrap, Microsoft’s robotic mosquito trap can distinguish one insect species from another. Leveraging machine learning technology, the trap can be programmed to distinguish between different species of mosquitoes. Ethan Jackson, a Microsoft researcher heads up Project Premonition, a research effort aimed at giving epidemiologists better tools for tracking disease outbreaks. While showing off how the high-tech trap works at the American Association for the Advancement of Science’s annual meeting in Boston, Jackson said that testing the traps last summer in Houston, Texas, showed they could determine the species successfully 80 percent of the time just through infrared scanning. Fine tuning the AI algorithm by for example, specifying the time of day and the amount of ambient light available allowed the accuracy rate to rise over 90 percent.
Research projects like the Global Virome Project are trying to find and sequence every virus that might threaten mankind. Animals harbor many diseases that infect humans, but it’s not practical to get blood samples from the entire animal kingdon.
“Then I realized that’s a mosquito’s full-time job,” Jackson said. “There are over 3,600 species, but if we can catch the right ones and do metagenetic sequencing on all the DNA and RNA we find in them, we can see the diseases on the move.”
While the first 30 prototypes, cost several thousand dollars each, Dr. Jackson hopes to get the price down below $300, so even poor countries where Ebola, malaria and yellow fever kill thousands of victims could afford them.
Each trap consists of 64 “smart cells,” compartments outfitted with an infrared light beam, like a sophisticated alarm system. When a mosquito of the desired species crosses the beam, its shadow changes the light intensity in a way that forms almost a fingerprint for that species, Jackson said.
Program the trap for the desired species —— and when one flies into a cell, its door snaps closed. In pilot testing in Harris County, Texas, last July and August, the trap was more than 90 percent accurate in identifying the insect buzzing through the door, Jackson said.
When each mosquito is captured like the Aedes aegypti mosquito that is the main Zika threat, sensors record the time, temperature, humidity and other factors, to show what environmental conditions have different species active. That’s information officials might use now to schedule pesticide spraying. In the future, the hope is rapid genetic scans of the mosquitoes’ blood can check for harmful pathogens — and can tell what animal the mosquito had been biting, Jackson said. If that work comes to fruition, he said the data may help predict emerging diseases.