Since early spring, Dr. Tom Karagiannis and his students and collaborators at Monash University in Melbourne, Australia, have been studying the SARS-CoV-2 main protease using Galileo and computing resources donated by Hypernet Labs. They aim to find antiviral treatments to fight COVID-19, which are the next best option in the absence of a working vaccine. Their first paper in a three-part series was published in the August 2020 issue of Computational Biology and Chemistry, and the second is currently under review.
The Karagiannis Research Group in epigenomic medicine ordinarily studies diseases such as diabetes, cancer, and Alzheimer’s. The pivot towards SARS-CoV-2 was relatively uncomplicated, though, because all of their work involves proteins and molecular modeling. Software tools such as GROMACS and AutoDock Vina (both of which can be easily deployed to the cloud using Galileo) allow researchers to simulate interactions between proteins and other compounds of interest. These compounds may be drugs, or they may originate in dietary sources. The Karagiannis lab simply turned these methods towards the SARS-CoV-2 main protease, a key protein responsible for the replication of the virus.
As soon as the COVID-19 crisis hit in early 2020, Dr. Karagiannis understood that the scientific community had a civic duty to respond, and to respond quickly. But one of the biggest obstacles to quick results in molecular modeling is a lack of adequate computing power. Normally, researchers and students run simulations on their laptops, studying a few compounds at a time, and it takes about two weeks to get results. With Galileo, Senior PhD candidate Julia Liang, master’s student Eleni Pitsillou, and other collaborating students were able to run the simulations overnight. “It’s easily ten times faster,” explains Dr. Karagiannis.
Liang adds: “Collaborative efforts are needed to tackle this pandemic, and tools to make this easier in any way are a bonus. I appreciate how accessible Galileo is when computing power can be a scarce resource. It’s great that it’s intuitive to use, allowing us to work with other students in a more efficient way.” Pitsillou, working closely with Liang, used cloud computing resources for the first time during this project. Happily, there was nothing new to learn. She found it incredibly easy to send her jobs to the Galileo computing stations and simply download results upon completion.
“Collaborative efforts are needed to tackle this pandemic, and tools to make this easier in any way are a bonus. I appreciate how accessible Galileo is when computing power can be a scarce resource. It’s great that it’s intuitive to use, allowing us to work with other students in a more efficient way.”
Computational methods, with the help of extremely powerful machines easily accessible through Galileo, allow pharmaceutical researchers to study hundreds, thousands, or even millions of compounds, which would be impossible to accomplish in a wet lab. It can take 3-6 months to synthesize a compound for experimentation.
Time is a challenge, but so is money. About half of the compounds in Dr. Karagiannis’ library of ligands–about 150 of them–have never been synthesized. Synthesizing them would cost about $3.2 million. For this project, he synthesized one compound for in vitro testing after running computer simulations. It cost $10,000 to create one gram.
“What computation allows us to do, which we can’t do in the lab, is we can look at, in our case, hundreds of compounds. Hundreds. Some people do in the hundreds of thousands. Some people do a million. So in a wet lab there’s no way. When we’ve got one or two leftover, we can do that in the lab. Maybe up to 10, if we’re lucky. But there’s no way we can do one million, for example,” says Dr. Karagiannis.
Galileo’s easy access to computing power is not only important for speed, but also for the accuracy or stringency of the tests. Higher stringency means observation of the protein-ligand molecular interaction over a longer simulated span of time. The protein and ligand are moving, so the longer time span allows the researcher to observe a greater number of configurations.
This is important because molecular dynamics is statistically based. Researchers need to understand the probability with which a drug molecule will actually bind to a protein as well as the probability that it will bind to the active site on the protein versus somewhere else. More observations result in greater accuracy. With the additional power accessible through Galileo, Dr. Karagiannis and his students could observe 2000 different configurations of the drug and protein, for example, rather than 150 configurations, and still get results overnight.
With Galileo, Dr. Karagiannis and his students could observe 2000 different configurations, instead of 150, and still get results overnight. They used to wait two weeks.
In the work for all three papers, the researchers used blind docking methods, which are especially computationally expensive. Instead of directing the ligand towards the active site to observe its behavior in a smaller space, blind docking involves observation of the entire protein, with the ligand moving freely. Blind docking, with the aid of computing power donated by Hypernet Labs, allowed researchers to ask, “If we run it 2000 times, where would these ligands end up, what percentage of the time?”
In the first two papers, Dr. Karagiannis and his collaborators focused on understanding the mechanisms by which drugs might bind and where they will bind. The “active site” on a protein is the exact location normally used by the protease to catalyze a reaction. In the case of the SARS-CoV-2 main protease, the active site is responsible for coding viral polypeptides to create functional viral proteins.
Ligands might also bind to “allosteric sites” and still have an effect on the active site and replication by affecting the shape and movement of the protein. It is possible to find these sites using the computationally expensive blind docking methods, rather than directing the ligand towards the active site.
In the third paper, the researchers turned their focus to the drugs. As part of this effort, they screened 300 compounds for potential effectiveness, narrowed the pool to 30, and then narrowed it again to 3. We asked, but Dr. Karagiannis is reluctant to release the names of the compounds too early, especially because some of them involve dietary sources. (Positive preliminary results could have a strong effect on public opinion, which can be misleading.)
Some of the compounds have made it beyond the computational stage and shown promise in in vitro tests. Testing in mice and in humans would normally still be needed, although it’s possible to work with humans immediately in the case of drug repurposing research and research on dietary compounds. Stay tuned for updates!