The IBM-built Summit has found a list of promising drugs for Covid-19
Covid-19 sickens people around the globe, scientists are rushing to find drugs that could help patients recover sooner. The never-before-seen pathogen can cause severe respiratory symptoms, including difficulty breathing and chest pain.
To aid in the search, scientists have enlisted the world’s most powerful supercomputer, the IBM-built Summit. Occupying the floor space of two tennis courts at the U.S Department of Energy’s Oak Ridge National Laboratory in Tennessee, the Summit can perform 200 quadrillion calculations each second — roughly a million times more computing power than the average laptop.
Last month, researchers used it to screen through a library of 8,000 known drug compounds to find those most likely to be effective against the coronavirus. The compounds included chemicals, herbal medicines, and natural products that have either been studied in humans or are already approved drugs — and, importantly, are already considered safe for humans. Summit narrowed down the dataset to a short list of 77 in just two days. Using regular computers, the process would have taken months.
“The logic is if any of those compounds works, it should be much quicker than the typical drug development process to get approval and widespread use,” Jeremy Smith, a molecular biophysicist at the University of Tennessee who ran the simulations, tells OneZero. He and his colleague posted their findings to the preprint server ChemRxiv in February and are updating the paper as they run more calculations.
If any of the compounds work in animals, scientists could skip the initial safety trial in people and go straight to testing drugs for their effectiveness in those who are sick.
Developing drugs is a notoriously lengthy process — it can take 10 years for a new medicine to reach the market from the time it’s discovered, and many fail because they’re not safe or just aren’t effective. That’s why supercomputers like the Summit are especially useful during a global outbreak of an infectious disease that has no known treatments.
To do the simulations, Smith used the virus’ genome, which Chinese researchers published to the web in January. The data revealed that the virus, now known as SARS-CoV-2, was similar to the coronavirus that causes severe acute respiratory syndrome, or SARS, and infects the body in a similar way. With this knowledge, they programmed the Summit to search for a very specific type of compound.
Coronaviruses get their name from the crownlike proteins on their surface, which allow the virus to bind to and infect human cells. The researchers used the Summit to pinpoint drugs capable of binding to these protein spikes in order to thwart the virus’ ability to get inside the body’s cells.
“I don’t know if any of them will work. Maybe no compounds in the database will work, or maybe several will.”
Complicating the process is the fact that these spikes constantly make lots of tiny movements. Researchers have to figure out how to model those movements to help find drugs that can work against them. “It’s a complicated mathematical problem,” David Turek, vice president of exascale systems for IBM, tells OneZero. (Exascale computing refers to the ability to make a billion billion calculations per second.)
A supercomputer can do this very quickly with machine learning algorithms. Using 4,608 nodes — the equivalent computing power of the same number of laptops — it takes a problem, chops it into pieces, assigns them to all of the individual nodes or computers, and then brings all those pieces back together to reconstitute the solution to the problem. It’s similar to a beehive, where a hundred or so different drones are working together for a common goal, but each one has its own mission. This capability allows researchers to perform incredibly complex tasks like drug discovery.
The results don’t mean the team has found a treatment or cure for SARS-CoV-2. The 77 compounds they identified still need to be tested in animals and human cells in a lab. Virologists at the University of Tennessee Health Center are beginning experiments now, but determining if any of the compounds are effective against the coronavirus could take months. That might not be soon enough to help patients right now, but an effective drug would be useful if the spread of the coronavirus lasts through next year or the virus becomes endemic, meaning it becomes a regularly occurring pathogen like the flu.
“I don’t know if any of them will work,” Smith says. “Maybe no compounds in the database will work, or maybe several will.”
In future outbreaks of new and untreatable diseases, Smith thinks these types of drug discovery simulations could be coordinated using the 500 or so supercomputers around the world so that scientists could jump-start drug testing. “We could have the scientific tools ready to go so that we could respond with the right science as quickly as possible,” he says.