In the realm of diagnostic tests for COVID-19, there are two main approaches: PCR tests, which detect the presence of the live virus; and serology tests, which detect antibodies that indicate whether someone has recovered from the disease.
But could there be a third way? Two companies in the Seattle region, Microsoft and Adaptive Biotechnologies, are on a quest to create a better diagnostic test.
The two Seattle-area companies are using machine learning to search for the unique signature associated with COVID-19 in the specialized cells that determine the human immune system’s response to the disease. Once that signature is identified, they say, it could lead to a new test that would identify the tell-tale signs of the disease in others, providing a new form of diagnosis.
As a machine learning challenge, it’s “a little bit of a scary problem,” said Peter Lee, Microsoft corporate vice president of AI and Research.
“It’s big,” Lee said. “It’s on the same rough scale as language translation, or topic identification of web documents. So there’s a scale to it that is a little bit daunting, but on the same scale as problems that we’ve actually solved at Microsoft.”
Based on their experience, the companies are optimistic about the prospect for developing a new type of test to help identify COVID-19 patients who are asymptomatic carriers of the disease; triage patients based on the likely severity of symptoms; and more effectively know if people have had the disease in the past.
One result could be a “true immune clearance to say, you’re now able to go back to work or go back to school, and we know that you’ve had the disease,” said Adaptive CEO Chad Robins.
The companies last week launched a virtual clinical study, seeking 1,000 people across the country who have been diagnosed with, exposed to, or recovered from COVID-19. The study is dubbed “ImmuneRACE,” for Immune Response Action to COVID-19 Events. It focuses on 20 major metro regions around the country, and requires participants to schedule a home visit by a mobile phlebotomist to collect blood samples and nose or throat swabs. As of earlier this week, the study had more than 100 participants. More information is available here.
Microsoft and Adaptive say they will make the data from the study freely available to other researchers working on COVID-19 initiatives
The initiative is part of a broader partnership between the companies, combining Adaptive’s immune science and Microsoft’s machine learning technologies to map the immune system’s response to a variety of diseases.
“Ultimately, what we’re trying to do is develop is a universal diagnostic (so that) when you walk into a doctor’s office, we can take a sample of a patient’s blood and tell you many diseases that patient has, all at the same time,” Robins said.
This approach also may result in “predictive power” for COVID-19, Lee said.
“Even if you haven’t been infected yet, we may be able to see something in the makeup of what’s called your T cell receptor repertoire … that may give indications as to why some people never get sick, why some people just get something like a three-day flu, whereas other people fall very seriously ill,” he said.
Robins and Lee discuss the initiative on this episode of GeekWire’s Health Tech Podcast. Listen above or subscribe in any podcast app.
