Revving Up COVID-19 Research

With the world racing to fight the coronavirus, Asia’s most powerful supercomputers are entering the fray.

AsianScientist (Oct. 13, 2020) – In the classic military treatise The Art of War, Sun Tzu emphasizes that a deep understanding of the enemy holds the key to their defeat. Though COVID-19’s battlegrounds look vastly different from ancient China—think sterile laboratories and hospital rooms instead of rugged terrain—the same principle still applies in the global fight against the coronavirus.

As soon as the outbreak’s extent became apparent, researchers worldwide were quick to react. Just a month after the first reported case in Wuhan, Chinese scientists had already released the genome of SARS-CoV-2—the virus behind COVID-19—allowing countries worldwide to swiftly create diagnostic kits.

Successfully developing interventions for COVID-19, however, involves a two-pronged approach: first, analyzing the biological molecules that equip the coronavirus with its deadly abilities. The molecules could then be used as potential targets for drugs or even vaccines. Second, making sense of the virus’ knack to stealthily spread in various environments. The thing is, these approaches take time. For instance, drug discovery and vaccine development both take years, if not decades.

With no time left to lose, scientists are hoping to accelerate the process with high-powered help. Across Asia, countries like Singapore, Japan and South Korea have launched special calls for proposals that leverage some of the world’s fastest supercomputers to fight the pandemic. Here’s a look at a few ongoing projects that could turn the tide in the battle against COVID-19.

Finding the weak spots

Over at A*STAR’s Institute of High Performance Computing in Singapore, Dr. Cheng Yuan is leading a team to investigate the structure of SARS-CoV-2’s main protease. Responsible for cutting precursors of viral proteins into functional pieces, the main protease plays a critical role in mediating viral replication and transcription, said Cheng. However, predicting a protein’s structure based on its sequence has always been a significant challenge. In addition, it’s entirely possible for the main protease to mutate—further complicating structural predictions.

“We are seeking artificial intelligence-assisted approaches to predict the main protease’s structure, taking into consideration any mutations,” explained Cheng.

Their approach will combine machine learning with multiscale modeling—a simulation strategy that simultaneously considers models at different scales of complexity. Given that proteins have four structural levels, she anticipates their experiments to be computationally intensive.

“The process is very demanding because of the large size of the data set and molecular simulations,” noted Cheng.

Accordingly, her team will be tapping upon the powerful computational resources of Singapore’s National Supercomputing Centre (NSCC). Specifically, they’ll be leveraging the flagship ASPIRE1 petascale supercomputer as well as an eight-GPU NVIDIA DGX-1 AI system with V100 cards and 13 petabytes of high performance storage.

Delineating the main protease’s structure would provide a deeper understanding of how it functions, and more intriguingly, give insights into its active site. When the active site is bound by an inhibitor, the protease is unable to function, interrupting viral replication and transcription. Just like striking an enemy’s weak spot in battle, knowing the main protease’s structure should aid in the design of drugs that inhibit its function.

Another SARS-CoV-2 protein being closely studied by researchers is the spike protein, recognizable as the distinctive spikes that dot the surface of the virus. The coronavirus uses the spike protein to bind to and invade human cells, through a receptor called ACE2. Similar to his counterparts in Singapore, Dr. Yuji Sugita from Japan’s RIKEN is seeking to predict the spike protein’s structure by simulating the way its atoms and molecules dynamically move over time. This technique, known as molecular dynamics (MD) simulation, would allow his team to discover structures that cannot be obtained through conventional means.

To achieve this, they’ll be running MD simulations on RIKEN’s Fugaku supercomputer, which is still in the process of installation. Despite this, even the partially available Fugaku is expected to run the simulations at a speed 125 times faster than its predecessor, the K supercomputer. By pinning down the complex structure of the spike protein, Sugita’s findings could help inform the development of drugs that block the interaction between the spike protein and the ACE2 receptor—preventing the virus from binding to the cells in the first place.

Driving drug discovery

Due to the pressing need for a COVID-19 treatment, alternative tactics are being used to shorten the time frame for drug discovery. To save time, rather than finding drugs completely new to science, scientists are repurposing drugs that have already been approved. Indeed, all treatments currently being tested in the World Health Organization’s global Solidarity trial are approved for use in other indications such as HIV or malaria.

To speed up the discovery process even further, a research team led by Professor Seo Sangjae of the Korea Institute of Science and Technology Information (KISTI) tapped on the world’s 14th fastest supercomputer, Nurion, to computationally screen thousands of drugs that could be given a second chance. Starting with a pool of almost 20,000 compounds sourced from the SWEETLEAD library and ChEMBL database, Seo and his team systematically evaluated the binding affinity of these compounds to the main protease of SARS-CoV-2 through a technique called molecular docking.

Comparable to finding the right key to a lock, the team then calculated the docking score by assessing which orientations of the compounds best fit the main protease’s active site and measuring the binding strength between the two molecules. Among those with the highest docking scores, the team chose 43 compounds to be investigated further using MD simulations. This motley set of compounds included antiviral drugs, antibiotics for pneumonia, vitamins, and drugs in clinical trials such as remdesivir and hydroxychloroquine.

“To identify drugs, it’s important to understand their interactions with enzymes,” remarked Seo. “Unlike other research projects that only performed molecular docking, we improved the accuracy of our results by conducting MD simulations as well.”

Publishing their preliminary results on ChemRxiv in early April, his team identified eight promising COVID-19 drug candidates that were all antivirals for either hepatitis C or HIV.

Incredibly, the team crunched all these calculations and simulations on Nurion in just one week.

“Each MD simulation was computationally demanding. To save time, we even had to run 43 drugs simultaneously. Harnessing supercomputing resources was the only way we could do this,” explained Seo.

Considering that Nurion is equipped with Intel Xeon Phi 7250 and 8,305 nodes, a study of a similar scale would require at least 200 days to finish on a personal computer.

Moving forward, Seo shared that the 43 drugs previously chosen for MD simulations are currently being experimentally tested in collaboration with KISTI’s partners. The experimental results will be compared with the MD simulations to help improve the computational process. At the same time, the team is now looking to make their docking calculations even more accurate by adding artificial intelligence into the mix.

Supercomputers have helped to model the circulation of fresh air (green) in a commuter train when the windows are closed (above) and when the windows are open (below). Credit: Makoto Tsubokura/RIKEN.

Profiling a killer

Aside from poring over SARS-CoV-2’s structure, it’s also equally important to consider how the virus is transmitted from person to person. After all, though the pandemic is nowhere close to ending, countries are already cautiously emerging from their respective lockdowns. As more and more people begin their much awaited return to the outside world, it is inevitable that we’ll be seeing fairly crowded spaces once again.

Unfortunately, crowds are also prime breeding grounds for the coronavirus. Human transmission is thought to mainly occur via droplet spread, when an infected individual releases thousands of droplets into the surrounding air via coughing, sneezing or even talking. But in a sinister twist, it’s been suggested that these droplets could be aerosolized into even smaller particles that could spread over larger distances and linger in the air.

Aerosol transmission poses a problem, especially in enclosed environments like public transport, most workplaces, classrooms and restaurants. To facilitate the safe resumption of activities that take place in these settings, Professor Makoto Tsubokura of RIKEN and Kobe University is looking to simulate the scattering of virus droplets under varying conditions on the Fugaku supercomputer.

Their project will consist of three steps, explained Tsubokura: first, assessing the infection risk of droplet and aerosol transmission in daily scenarios, including commuting and offices. Based on their results, his team then hopes to propose immediate countermeasures that could reduce infection risk—ranging from opening or closing windows to the strategic placement of partitions. Finally, they seek to suggest long-term measures in the fight against COVID-19, such as improved air conditioning or ventilation systems.

“We are required to produce many results in a very short period of time,” said Tsubokura. “In addition, in cases like train simulations, we have to consider very crowded cabins with more than 200 passengers, running at 80 kilometers per hour.”

Given that the team has to evaluate the ventilation effect of open windows in such complicated scenarios, Fugaku’s massive computational resources are required for their simulations. His team’s academic and industry partners will then go on to experimentally validate their results.

From the simulations, Tsubokura is also aiming to create accessible animations of the virus’ droplet and aerosol spread so that the public can easily understand the risk of infection and the need for countermeasures.

“People don’t understand COVID-19 simply because it is invisible. Our goal is for our HPC results to help people to be more informed, especially government planners who will establish guidelines for a post-COVID society,” he added.

As the pandemic continues to rapidly unfold, it may be hard to see the light at the end of the tunnel. But the coronavirus may have finally met its match in the form of supercomputers like ASPIRE1, Fugaku and Nurion. With the world’s best minds and fastest supercomputers joining forces to attack COVID-19 on all fronts, there’s still hope yet.

This article was first published in the print version of Supercomputing Asia, July 2020.
Click here to subscribe to Asian Scientist Magazine in print.


Copyright: Asian Scientist Magazine.
Disclaimer: This article does not necessarily reflect the views of AsianScientist or its staff.

A molecular biologist by training, Kami Navarro left the sterile walls of the laboratory to pursue a Master of Science Communication from the Australian National University. Kami is the former science editor at Asian Scientist Magazine.

Related Stories from Asian Scientist