Supercomputing and big data have taken yet another step forward in leading to new treatment for cancer.
Schrödinger, a company that offers molecular-modeling and drug-design software, and Nimbus Discovery are using Cycle Computing’s 50,000-core supercomputer to run a virtual screen to find a protein target responsible for cancer.
Schrödinger helps fund the work of Nimbus, which performs drug discovery and molecular modeling.
Cycle Computing announced the creation of the supercomputer cluster, code-named Naga, at the Amazon Web Services summit in New York City on April 19.
AWS’ cluster of servers power the research in a cloud environment. The CycleCloud high-performance computing (HPC) software brought the cluster to the supercomputing level.
“We take those servers and turn them into a working, functioning supercomputer,” Jason Stowe, founder and CEO of Cycle Computing, told eWEEK.
Schrödinger offers Glide, a computational docking application that performs the screening of compound libraries as researchers try to identify proteins that may have an effect on cancer activity. These discoveries can lead to the development of potential treatment.
“What they’re trying to do is find molecules that would fit into the target on this protein, much like a lock-and-key kind of scenario,” said Stowe.
As a mapping application, Glide allowed researchers to virtually screen 21 million molecule conformations against a possible cancer target. The software simulates the placement of a molecule into the protein, said Stowe.
“In order to do that in a reasonable period of time, they need multiple machines, and that’s essentially what a high-performance-supercomputing cluster is used for,” Stowe explained.
Following computational testing, Schrödinger and Nimbus are physically testing the molecules in a lab.
“We will be purchasing and assaying the compounds that came from this virtual screen to confirm that indeed better science gives better results,” Ramy Farid, president of Schrödinger, told eWEEK in an email. “It’s hard to imagine that this will not be the case, but we have to do the experiment to confirm it.”
Researchers also used the supercomputing vendor’s CycleServer software to perform analytics, diagnose performance and manage the scientific workflow for the project.
Researchers completed the run in less than three hours compared with traditional methods, Stowe noted.
“Essentially, they’re taking months off of this process,” he said. “If you were to try and run this [in-house], you would either miss drugs because you did it the old way or essentially, you’d wait considerable periods of timea couple of months to get the results backand then have to sift through them.”
The computations cost $4,900 per hour at peak without up-front capital to access the computing capabilities through the cloud, Cycle reported in a blog post. An in-house supercomputer would cost tens of millions of dollars to power, cool and manage, according to Cycle.
“It would take a considerably longer period of time, so they’d never be able to run it in house,” said Stowe.
By combining Cycle Computing’s supercomputing software with AWS servers, researchers are able to perform complex computations in a much shorter time for just a fraction of what the HPC infrastructure costs, Terry Wise, director of business development at AWS, said in a statement.
When performing drug research using supercomputing, scientists usually have to weigh “trade-offs” between time and accuracy, according to Schrödinger’s Farid.
“This virtual screen identified a number of compounds that were missed in the screen that we did on the same target on our in-house cluster, which needless to say is much smaller than the 50,000-core Amazon cluster,” said Farid.
Cycle’s supercomputing allowed Schrödinger and Nimbus to run the virtual screen with the proper levels of scoring and sampling, Farid noted.
In the past, only the largest companies could afford 50,000-core computing, said Stowe. He noted that Schrödinger has only 100 to 150 employees and that it’s uncommon for a pharmaceutical company to deploy 50,000 cores for high-performance computing.
Using supercomputing could allow more drugs to pass clinical trials and get to market, Farid suggested.
“A majority of projects never even get to the development candidate stage, and for the ones that do, a large majority fail at some stage in the clinical trials and therefore never become a marketing drug,” said Farid. Faster, larger computers can “almost certainly” solve this problem, he said.