REDMOND — A groundbreaking system that would use artificial intelligence to diagnose several diseases from a single blood test may not be far off.
The project, a collaboration between Microsoft and Seattle-based Adaptive Biotechnologies, could start diagnosing conditions “within a small number of years,” said Peter Lee, corporate vice president of Microsoft Healthcare, speaking at a conference for healthcare developers at Microsoft headquarters here Tuesday.
Lee’s optimism has come along way since he first heard the idea.
“This sounds a lot like science fiction. And in fact, when the idea was first presented to me by the people at Adaptive Biotechnologies, I was also somewhat skeptical about this, despite the fact that that I’m a big fan of Star Trek,” he said in a reference to the Tricorder, a handheld diagnostic device made famous by the Star Trek franchise.
Lee became convinced of the idea after seeing how Adaptive was able to label 100 billion data points per month. “The prediction is that, within a small number of years, we might have enough training data, if fed to a suitably robust cloud-scale machine learning algorithm, to actually achieve some diagnostic results,” Lee said
The system works by interpreting information about the immune system that can be read from a small vial of blood. Adaptive’s technology decodes the genetic information that is associated with T cells, which are vital to immune response. Inside those cells exists “all of the information that’s needed to determine exactly what cancer, what infectious diseases and what autoimmune disorders your body might be coping with,” Lee said.
The partnership between Microsoft and Adaptive was first unveiled more than a year ago. In January, the companies announced that the AI system is up and running and will initially focus on diagnosing type 1 diabetes, celiac disease, ovarian cancer, pancreatic cancer and Lyme disease.
As part of the deal, Microsoft invested $45 million in Adaptive, and Adaptive committed to spending a minimum of $12 million on Azure cloud services over seven years.
But in order to diagnose a disease accurately, the companies will have to do more than read the immune system. They’ll have to know how to interpret it. “The problem is, we don’t exactly know how to extract this information,” Lee said.
Adaptive recently filed for an IPO and in doing so, warned that the work with Microsoft might not go so smoothly.
“Our collaboration with Microsoft is in the early stages, and our computations and algorithmic-based methods are largely untested,” the company said, listing the cautionary risk factors in its filing for the public offering. Adaptive added that the work “may not yield clinically actionable insights on a timetable that is commercially viable, or at all.”
But if the teams are able to spot disease information from these immune system readings, it might lead the way to a “completely universal disease diagnostic, a simple blood test that you could take once or twice a year, that would give you an early diagnosis of infectious disease, cancer and autoimmune disorders,” Lee said. The goal is to do this all within a few hours at a reasonable cost.
Microsoft’s cloud business stands to profit from a world in which diagnostics shift from laboratories to data centers, given the massive amount of computing power involved. “These precision medicine ideas, these genomics, immunomics and proteomics-based ideas involve gigantic data workloads, data volumes and data complexities that really defy human comprehension,” Lee said.