Using supercomputers and artificial intelligence to fight COVID-19 at UW Bothell

SEATTLE — A new program called “DeepTracer” developed by a team of University of Washington scientists uses artificial intelligence to model the atomic structure of COVID-19.

The program uses machine-learning technology that could help design vaccines and treatments for the virus and it's already being used by scientists around the world.

We’ve all seen the iconic image of SARS-CoV-2 - a ball with red spikes. But the image is an artist’s illustration.

Under a very powerful microscope, one of those red spikes (a spike protein) looks like a gray jagged blob.

“You only see it from the outside, you see the volume. But you don’t know the detail inside,” said Dong Si, an assistant professor at the University of Washington Bothell School of STEM. His team used machine learning and AI to figure out what the inside of a COVID virus particle looks like.

“The pharmaceuticals need accurate and enough detailed structural information in order to design vaccines,” Si said. “This spike protein is just a little tiny - one piece of this red thing,” he said, pointing to one of the red pieces jutting out from the artist illustration. “But it will attach to the human cell so it’s very important,” he said.

DeepTracer uses a supercomputer to predict the structure inside viruses down to every atom.

It takes an image from a gray mass to showing the RNA and atomic structure inside.

The team used pictures and atomic mappings from old, known viruses to teach the computer to figure out what the inside of a virus looks like, based on the outside.

That means the information is helpful not only for vaccine development but also possible COVID treatments.

“If you only know the rough shape, it’s hard to design an accurate chemical drug,” Si said.

Dan Bustillos of the School of Nursing and Health Studies at UW Bothell says viruses mutate and change often. COVID has already mutated once, at least.

If that happens again, he says this technology will be able to save precious time and likely lives.

“The AI research into the protein structure of viruses is so vital because we can get a very accurate snapshot of what a particular virus is, and make vaccines that adapt well to that virus and nullify its effects,” Bustillos said.

DeepTracer predictions take only minutes. Manually, that modeling work can take weeks or even years.

“COVID-19 won’t be the last pandemic that we face,” Bustillos said. “It’s just another arrow in the quiver of science, to be able to speed up time between the emergence of a new infectious disease and the availability of a vaccine for it,” he said.

The accuracy of DeepTracer’s predictive modeling averages 85% right now and needs to be manually modeled further, but gets a big chunk of the work done quickly. Si said they are working towards 100 percent accuracy.

The program is freely available for researchers to use around the world.