AI’s Accelerating Our Race Towards a COVID-19 CureJune 3, 2020 - 8 minutes read
Although most of us are confined to our homes, scientists are still working around the clock to find a cure for the coronavirus, and they’re working faster than ever. The COVID-19 pandemic is speeding up research all over the world.
Neural networks are watching incoming data for any indication of a potential cure, and the results are promising. At Argonne National Laboratories, one of the nine supercomputing systems managed by the U.S. Department of Energy, researchers are modeling how existing drugs “dock” with viral proteins.
This research focuses on the potential strength of an attachment of the drug to the functional protein of the virus, effectively rendering the virus useless. Using an innovative computer chip from San Francisco-based Cerebras Systems, the supercomputer is working long hours to uncover information that could make the difference between finding a cure tomorrow or years from now.
Drug Discovery with AI
The model at Argonne Labs is a medical application that uses machine learning to predict docking scores. These scores help scientists prioritize shortlist drugs to experiment with at the lab.
Argonne Labs is a high-energy supercomputing lab that already works overtime in its research endeavors. But this project is unlike any other; its speed and efficacy affect how many lives we can save before the virus spreads further.
Rick Stevens is the associate laboratory director at Argonne. He says that the lab is working so fast that it’s accomplishing feats that would normally take years within just months. The lab set up three different neural networks to calculate a combined score, rather than relying on a score from one algorithm.
However, for Stevens, that’s not the part that’s most interesting; he’s more fascinated by the fact that the models use images of molecules to simulate protein docking, whereas most other researchers are using actual chemical models of the molecules.
The team isn’t sure exactly why this works so well, but they’re publishing several papers to accelerate innovation in the scientific community.
Better Images = Speedier Success
Neural networks are a subset of deep learning, which is a subset of machine learning development. They make graphs to illustrate connections between inputs, and this helps elucidate relationships for the researchers.
280 scientists across 20 labs and research facilities are working on finding the cure for coronavirus in this case. This includes researchers from London to Chicago to San Diego.
The key to the massive effort and rapid research is Cerebras’s innovative computer chip that’s been deemed to be the largest chip in the world. Argonne is the first customer to use the chip, named CS-1; it allows the researchers to iterate on many different combinations of neural networks without needing additional CPUs or GPUs.
It goes without saying that the new chip can handle tons more images than any of the chips out on the market right now. On a standard GPU, memory fills up quickly, especially when you’re training the algorithms on high-quality images. But this chip allows many more high-resolution images and maintains its speed and processing power at the same time.
Andrew Feldman is the CEO and a co-founder of Cerebras. He says the chip was built for this type of work.
“They’re running 30, 50 days on machines that cost a quarter of a billion dollars to do the work that we’re doing in a single machine that’s the size of a dorm fridge. And that’s enormously exciting,” Feldman says.
Finally Finding a Cure
Argonne has pinpointed several molecules that are showing inhibition, which is a great sign for a molecule to be a potential cure. However, other than the computer processing work, there are many more steps that have to be taken before the promising molecules can become part of a vaccine.
For example, the compounds must be tested in the lab with live virus assays first. If they pass the test there, they will be testing in animals. If a molecule isn’t already an existing drug, chemists will need to take the time to synthesize the molecule. And that’s why it can take 18 months to 10 years to develop a vaccine for a novel virus.
Stevens says, “We’re trying to validate whether the computational work actually holds up in the experimental work, and a number of those are progressing to wholesale assays to test” for efficacy against the virus.
Argonne is working on other projects that could impact the virus and its cure, like understanding the virus’s protein structure, looking for antibodies using machine learning, and searching for the right binding agent in humans. This work is tedious and complex, but it could help us answer any questions that arise during the vaccine development process without wasting more time.
Tomorrow’s Good News
Stevens says the teams are working on publishing a few papers about the process, the new chip, and the imaging neural networks, which should be available for reading in the next couple of weeks. He is ensuring the data is properly peer-reviewed and tested before publishing the results, saying that he doesn’t want to publish something that hasn’t been vetted or tested extensively.
Until we’ve got the vaccine, researchers and scientists all over the world are going to be sacrificing sleep to find promising molecules and compounds. How’s the virus impacting you, and what hope do you have for a vaccine? Let us know in the comments below!Tags: AI app developer San Francisco, app developers san francisco, artificial intelligence app development, eHealth app development San Francisco, machine learning app developer, machine learning applications, machine learning apps, medical app developer, MedTech app development San Francisco, mobile app developers San Francisco, san francisco AI app developer, San Francisco eHealth app developer, San Francisco MedTech app development, San Francisco mobile app developer