Microsoft Premonition
Established: March 2, 2015
Our health depends on timely data about the pathogens around us
Disease epidemics have impacted every society and economy in human history. The key to reducing future epidemics is the early detection of potential pathogens — before they cause large disease outbreaks. This gives researchers time to develop new treatments, public health organizations time to prepare responses, and individuals time to minimize their exposure to sources of disease risk.
However, detecting pathogens before they cause outbreaks is no easy task. Pathogens move through the environment in complex ways that are difficult to monitor by traditional methods. It is estimated that 60 – 75% of emerging infectious diseases are caused by pathogens that jump from animals to people. Viruses like Zika, dengue and West Nile move between humans, animals, and mosquitoes in complex cycles. Yet, today we have limited technologies and capacity to monitor potential pathogens as they move through the environment.
The goal of Microsoft Premonition is scalable monitoring of the environment to detect disease threats early, using robotics and genomics. Our robotic smart traps continuously monitor the environment for important types of insects, such as mosquitoes, which both transmit pathogens and collect blood samples from other animals. Meanwhile, our cloud-scale genomic analyses try to identify all the species of organisms and viruses in environmental samples to spot new transmission patterns.
What if we could predict epidemics like we predict the weather?
Collecting disease transmitting mosquitoes with robots
Most terrestrial animals are arthropods (e.g. insects, spiders, and ticks) estimated at millions of species. Arthropods are essential for our ecosystems (as crop pollinators, food for larger animals, predators of smaller arthropods, and natural recyclers). However, arthropods such as mosquitoes are some of the most important transmitters of pathogens that cause human disease. Monitoring mosquito populations and predicting their distributions is essential for predicting future disease outbreaks.
An array of robotic designs being tested on live mosquitoes at Microsoft Headquarters
Microsoft Premonition’s robotic smart traps are designed to adaptively lure, identify, and selectively capture targeted mosquito species in the environment. We believe robotics has the potential to dramatically scale monitoring of these major disease carriers. Because there are thousands of known mosquito species, we are exploring state-of-the-art AI-based species identification, developing novel robotic designs, and building unique capabilities to evaluate designs on many mosquito species.
Working with partners in real environments
Developing scalable monitoring solutions for real-world is an interdisciplinary effort requiring a diverse set of partners. Microsoft Premonition embraces academic, governmental, and industrial partners to help deploy and evaluate our technologies in complex ecosystems. We have focused our field deployments on answering several related scientific questions: (1) How many types of arthropods will visit a robot, and how hard is it to autonomously classify them? (2) Can autonomously collected data be used to build better forecasts of disease risks, and can those forecasts be used to better inform human health programs? (3) How reliably can microbes, viruses, and other environmental DNA be recovered from robotically collected specimens in urban and rural environments?
For example, in a collaboration with Harris County Public Health, we trialed our technologies in Houston, TX during the peak of Zika transmission risk in 2016. We saw that robot field biologists could be trained to identify and selectively capture medically relevant mosquitoes with high accuracy (> 90%). Our robots were also able to digitize mosquito behaviors at high resolution, allowing us to better understand how they moved through the environment. Our genomics analyses were able to detect microorganisms and viruses in mosquito specimens, and identify the types of animals on which they fed. Since then, we have explored diverse habitats ranging from the southern tip of Florida to the remote forests of Tanzania. Along the way, we have continued to learn how these technologies and data sets can assist our partners with their essential human health missions.
Detecting potential pathogens with DNA sequencing
Monitoring the environment for pathogens is difficult, because detecting microorganisms and viruses usually requires targeted chemical tests (or arrays of chemical tests). It may be unclear which tests to run on a sample, or targeted tests may not exist. Fortunately, DNA sequencing opens the door to new broad-spectrum approaches. By converting the genetic material of an environmental sample into digital data, we can computationally scan a sample using trillions of genetic signatures to discern the viruses, bacteria, fungi, and higher-order life in a sample.
Finding an unexpected viral signal in a large mosquito metagenomic dataset
Microsoft Premonition’s metagenomics pipeline estimates the organisms and viruses in a sequenced environmental sample using cloud-scale machine learning and a large database of reference genomes spanning the tree of life. This technique allows us to test for all known sequenced pathogens and also detect related viruses and bacteria. Because we also scan for higher-order life, we can look for complex associations between potential pathogens and their possible hosts. For example: does an unknown virus tend to appear in mosquitoes that have evidence of feeding on both birds and humans?