How Supercomputers Can Help Answer Questions About The Coronavirus

May 4, 2020

There’s a lot we don’t know about the novel coronavirus. How many virus particles does it take to cause infection? Why do some people develop severe infections while others are asymptomatic carriers? What drugs might work as therapies for COVID-19?

Supercomputers may be able to help answer some of these basic questions.

The Pittsburgh Supercomputing Center has announced that it’s offering up its specialized molecular dynamics supercomputer, called Anton 2, to research teams as part of the COVID-19 High Performance Computing Consortium.

Applications are now open for non-commercial research teams to run simulations on the machine, which was developed by New York City-based D.E. Shaw Research.

Anton 2 is specifically designed to quickly model interactions between biological molecules, such as the novel coronavirus and the specialized lung cells responsible for oxygenating the bloodstream.

The biological phenomena each simulation aims to illuminate are incredibly tiny, said Philip Blood, senior director of research at Pittsburgh Supercomputing Center. For example, researchers have mapped the “spike protein” present on outside the SARS-CoV-2 virus cell. This protein attaches to a protein found in human airway cells, and that’s what allows the virus to infect them. Anton 2 could help researchers understand the exact mechanism by which the virus’s spike protein binds to the airway cell protein.

“It’s like a computational microscope,” said Blood.

While these interactions in nature happen almost instantaneously, modeling them on the computer can take days or weeks. Every interaction is broken down into steps, each representing a tiny fraction of a second, called a femtosecond, which is one quadrillionth of a second. The machine calculates how each atom of each molecule will interact in that time frame, based on the laws of physics. A typical simulation might have around 100,000 different atoms, according to Blood.

“[We can] see at this atomistic level of detail what is happening with these molecules when they interact. How are they changing, how they are interacting, what are the key parts of the molecule that determine its motion and therefore the function that it’s able to fulfill,” he said.

It takes billions of femtoseconds for anything “biologically interesting to happen” said Blood. In biological time, that’s still basically instantaneous; a billion femtoseconds is a microsecond, or one millionth of a second. Simulating 20 microseconds of biological time on Anton 2 takes a full day. While that might not seem terribly fast, Blood said a traditional supercomputer could take 10 or even 100 times as long to do the same work.

“If we could simulate life in real time, we’d have it made,” Blood said. “It’s very computationally intense, so it takes much, much longer.”

The idea behind Anton 2 is not to move faster than nature; it’s to help drive research forward faster than lab experimentation can on its own.

“The simulations can guide and point experimentalists in a good direction, and then in turn the experiments can suggest new simulations that can get at information that it’s hard for experimentalists to get at,” Blood said. “So it’s this working together with experimentalists, back and forth, suggesting new directions, using the strengths of simulation and the strengths of experimental work."