A modern twist on a World War II combat method is aiding the fight against the deadly Ebola virus — and could help avoid a catastrophic influenza pandemic.
Eric Jakobsson from the University of Illinois at Urbana-Champaign and the National Center for Supercomputing Applications, and Amir Barati Farimani from Stanford University, took their inspiration from the Blitz, and the randomly determined automatic firing patterns of anti-aircraft guns to protect London from German war planes. They’ve used similar stochastic algorithms, powered by GPUs, to develop multiple simulated molecular models that predict which antibodies would best combat certain strains of Ebola.
Keeping up with the evolution of Ebola, which mutates every three to four years, is difficult. Trying to outmaneuver influenza, which mutates every three to four months, is another story. Yet that’s just the challenge the research team intends to take on after Ebola.
“We know that if we continue on the path that we’re on, we will eventually have an influenza pandemic that will kill millions and millions of people. It’s inevitable,” said Jakobsson.
“The virus will eventually mutate in a way we can’t anticipate,” he said. “We’re hoping that the methods we’ve developed for Ebola can be applied to the flu to respond rapidly enough to a new super-virulent strain.”
Jakobsson and Farimani, a postdoctoral research fellow in chemistry at Stanford, set out to combat Ebola with a combination of history and technology. With the WWII history providing the strategy, the pair turned to the NCSA’s Blue Waters supercomputer, running NVIDIA Tesla K20X GPU accelerators, to run Farimani’s molecular simulations.
Speeding Up Flu Virus Research
Normally, it would take 10 to 15 years and thousands of people to find an antibody to counter Ebola as medical researchers wait to see reactions in human test subjects.
By mimicking that process in simulations on Blue Waters, and applying bioinformatics and heavy data analytics, Jakobsson and Farimani were able to shorten the process of designing antibodies effective against Ebola.
The researchers ran hundreds of simulations, each of which required 24 to 48 hours of compute time. Ultimately, they predicted the movement of the modeled Ebola strains for about two years into the future.
“That creates a huge space for us to design the next generation of antibodies to counterattack that,” said Farimani.
Had the simulations run on CPUs, they would have taken up to 100 times as long, said Farimani. It’s that superior performance that has made a GPU-centric approach to solving the problem a no-brainer.
“That gives us the opportunity to try more and more,” said Jakobsson. “That’s the gift of GPUs.”
Quickness of Response Could be Difference
As the team works on getting ahead of influenza, the computing demands will ramp up, necessitating the use of machine learning and deep learning techniques, Jakobsson said.
“The great majority of flu strains cause only relatively minor illness and only a few people die, yet there is the potential for a really horrible pandemic,” said Jakobsson. “We want to help humankind be prepared to deal with that.”
Jakobsson and Farimani are preparing a research paper that will illustrate how their work demonstrates the promise of computation design of antibodies to reduce trial and error and thus enable quicker responses. When a truly virulent influenza strain arrives, they say, the ability to respond swiftly will be essential.
Feature image credit: NIAID