- Opioid addiction is ravaging large parts of the US
- Oak Ridge National Laboratory used the fastest supercomputer in the world to study the problem
- This research could extend beyond addiction to other urgent challenges
The opioid crisis in the US is out of control. According to the Centers for Disease Control and Prevention (CDC), opioid overdose deaths in 2017 were 6 times higher than in 1999. Around 130 Americans lose their lives every day to illegal and legal opioids.
As with most major public health crises, there are multiple causes behind this epidemic.
The importance of this research hasn’t been lost on the scientific community, as the ORNL team was awarded the 2018 ACM Gordon Bell Prize for their work. Science Node sat down with Daniel Jacobson, a chief scientist for Computational Systems Biology at ORNL, during the SC18 conference in Dallas, Texas, to learn more.
Digging through data
Opioid addiction can easily sneak up on a person if they aren’t careful. A small dose to help with pain can spiral into heavy use as a person’s body builds tolerance. Before long, someone who was legally prescribed pain medication may be looking to score illegal drugs just to get the same high.
However, Jacobson points out that the problem often starts even before a person is prescribed medication. In fact, there seems to be a genetic component for how people feel pain.
“One of the issues with addictions and substance abuse disorders is, one way or another, they're a palliative response to pain,” says Jacobson. “For opioids specifically, problems often come when they're being used to treat chronic pain. But different people have very different pain thresholds and very different propensities to having an injury result in chronic pain. We want to understand the genetic architecture of that as well.”
To study this problem, the ORNL team needed to comb through large amounts of genomic data. The answer was the CoMet algorithm, which is 3-4 orders of magnitude more efficient than other state-of-the-art methods. What’s more, Summit’s computational power is making this project much easier. This machine’s capabilities, combined with CoMet, enabled the analysis of population sizes exceeding 4 million.
And then there’s the complexity of the cells themselves. A big issue was combinatorial complexity, which is the difficulty of studying combinations of observable objects.
“The combinatorial complexity of a single cell, whether it's a plant or a human, is you look at how to discover these higher order combinations of interacting parts,” says Jacobson. “The possible combinatorial space is 10 to the 170. A little bit of context: the number of atoms in the observable universe is thought to be about 10 to the 83. We'll never be able to brute-force and look at all the possible combinations.”
With brute-force out of the question, the CoMet algorithm was a new way to break down this combinatorial complexity. By using mixed-precision arithmetic, CoMet became the first application to attain exascale speeds.
This degree of difficulty points to the need for a machine like Summit. With a performance of 143.5 petaFLOPS, Summit easily snagged the top spot on the TOP500 list of most powerful supercomputers. Designed by IBM, Summit has an entirely new computing architecture that combines IBM POWER9 CPUs with AI-optimized NVIDIA GPUs, all linked at extremely high speeds.
“The partnership began in 2014,” says Jacobson. “The P9s are very powerful with capable CPUs and there's the NVIDIA NVLink interconnect between the CPUs and the GPUs. Basically, the CPUs on the compute node do some of the computations and orchestrate the work on the GPUs. This whole package working together was what enabled us to be successful.”
Preventing is better than fixing
While science is often rewarding in its own right, the work being done by ORNL could have important future applications. Specifically, this kind of research could help us understand who is most at risk for addiction.
“Often with addiction, the more we understand chronic pain, the better therapies we can design that are not addictive,” says Jacobson. “Even just knowing the underlying genetic architecture means you can develop a blood test to be used in the clinic to say, 'Guess what? This person's really likely to be addicted to opioids if you prescribe them, so don't prescribe them. Let's look for another method.’ So that's just preventive predictive care.”
That said, scientific study of one area generally allows for better understanding of others. While Jacobson and the rest of the team want to know more about opioid addiction, they’re also open to how this research could affect different endeavors.
“On the bioenergy side, we're doing a lot of work with microbiomes and how microbiomes and pathogens interact with plants,” says Jacobson. “We're doing a lot of work in sustainability. How does the system cope with different types of stress, whether that's drought or climactic adaptation across the native range of a species? On the human side, Alzheimer's is very high on our list, as are a lot of neurological disorders.”
Opioids are sometimes a necessary tool for getting a pain-stricken person back on their feet. However, the recent epidemic shows that some people are entirely too susceptible to the drugs’ powers. With more time, hopefully scientists like Jacobson can help create treatments that help more than they hurt.