IBM-Oxford Team Uses Supercomputers to Design New Drugs Against COVID-19

Inside IBM Research
3 min readNov 2, 2020

By Katia Moskvitch

With the second wave of COVID-19 gaining strength, researchers are in a race against time to find a treatment or a vaccine.

One international team of scientists from IBM Research and Oxford University is trying to design molecules that would interfere with the molecular machinery of coronavirus, the virus that triggers the disease. If successful, such molecules could become the basis of a new drug to treat or slow COVID-19 infections.

“We are blending techniques such as advanced machine learning, computer modelling and experimental measurements to accelerate the discovery of these new molecules,” says the lead researcher Jason Crain, IBM Research physicist and visiting professor at the University of Oxford. He details his team’s work in a recent COVID-19 High-Performance Computing Consortium’s webinar.

It’s still early days — the team is only four months into the project — but the researchers have already identified several compounds that look promising based on the computational modelling. The scientists now have to test them in a lab, says Crain, and the experiments will take several weeks.

While the ongoing COVID-19-related work is new, Crain’s team has been for many years working on drug discovery, most recently in the area of antibiotic resistance. “We pivoted this earlier work, quickly adapting some of the fundamental methods we had previously developed, to address COVID-19,” Crain says.

The biggest challenge for the team, just like for any other team searching for a new drug to halt the pandemic, is dealing with an immensely vast chemical space within which to identify new functional compounds. To address it, the researchers are combining cutting-edge AI methods with modelling on two supercomputers offered by the COVID-19 HPC Consortium — IBM Summit at Oak Ridge National Laboratory and Frontera at the Texas Advanced Computing Center.

Without these extra computing resources, Crain says, “the throughput of the computational screening stages would have been prohibitively slow.” After all, the computational modelling of a myriad of AI-generated candidate compounds is among the most demanding and time-consuming steps in the discovery pathway.

Computer modelling on Summit and Frontera has allowed the team to screen compounds and reveal their mode of action at the molecular scale, so that they have to synthesize and test experimentally only the most promising ones. “Summit and Frontera allow us to perform calculations of how candidate drug molecules bind to viral proteins much faster than would have been possible otherwise,” says Crain. “The Consortium resources have allowed us to incorporate very HPC-intensive steps into the screening protocol, which is a very powerful approach but rarely possible to do.”

The Consortium has also helped, says Crain, to bring together an international team of experts. “Some of the Oxford team, for example, have extensive experience in the structure of viral proteins, and techniques related to screening of candidate drugs,” he says. “The AI teams at IBM in New York and in the UK have been working on developing new methods that can ‘discover’ functional molecules — which may or may not have been made previously — very efficiently.”

This article first appeared on the COVID-19 HPC Consortium blog