The biopharma industry’s adoption of artificial intelligence has exploded in the past few years, giving researchers access to computing power orders of magnitude greater than what was possible when the world first faced the novel coronavirus’s older sibling, the bug behind the 2003 SARS outbreak, or even during Ebola’s spread as recently as six years ago.
But when it comes to COVID-19, where do researchers start?
Currently, AI is very good at finding patterns within very large data sets, according to Arvind Ramanathan, a technology leader at the U.S. Department of Energy’s Argonne National Laboratory, which is providing supercomputing resources to a public-private consortium of COVID-19 researchers.
“AI is especially good in terms of helping our understanding with what we consider to be data that is not fully characterized...AI is very good at picking up on those patterns,” Ramanathan said during a Fierce AI Week panel discussion. “We’re in a very good position in terms of exploiting AI to understand quantitatively what is there in the data."
The Argonne lab has been using AI to build dynamic, physics-based models of proteins and how they fluctuate and interact within cells, to help uncover potential drug targets. Subsequent chemistry studies can generate data fed back into the AI model, creating a tight loop between experimental and computational work, Ramanathan said.
From there, drug designers have been presented with two paths: to use AI to sift through tens of thousands of potential therapies already on the books for a possible match, or to create something wholly new. One offers a shorter path to the public, while the other may provide a more complete solution.
BenevolentAI has opted for the former. Earlier this year it identified baricitinib, also known as Eli Lilly and Incyte’s rheumatoid arthritis drug Olumiant, as a potential therapy for COVID-19, and results from a phase 3 trial are expected soon.
“We figured if we were going to help in this pandemic, that we had to find an approved drug that we could repurpose, so that it could be used straight away,” said Peter Richardson, BenevolentAI’s VP of pharmacology.
“For all the elegance of finding new targets and new drugs, the problem is, with a virus that’s spreading at this speed, the regulatory and clinical trial hurdles are just too long and high for us to be able to do much at least in the first year,” Richardson said.
RELATED: Eli Lilly, spurred by AI analysis, starts pivotal study of arthritis med Olumiant in COVID-19
Baricitinib works against the novel coronavirus in two ways: It inhibits some of the kinases that enable the virus to enter human cells, and it helps dampen the immune system overreactions seen in the most severe COVID-19 cases.
“So potentially we have an antiviral and an anti-inflammatory combined,” Richardson said. BenevolentAI’s systems also found that baricitinib’s movement through the liver and kidneys meant it could be safely paired up with other, direct-acting antivirals—such as Gilead’s remdesivir, with one combination currently being studied in an NIH-led trial.
“The sum of human biomedical knowledge has a lot of gems in there, because no one person can handle the enormous amounts of data,” he said. “Applying AI to that allows us to identify relationships that would perhaps never be recognized.”
RELATED: Insilico's AI networks generate custom lead compounds for fibrosis in less than 50 days
Meanwhile, companies such as Insilico Medicine aim to use AI to create a purpose-built molecule for stopping COVID-19, and it’s already seeing early preclinical successes.
“AI can be used at every step of drug discovery, drug development and personalized medicine, and also demographic analysis in tracking COVID-19,” Insilico founder and CEO Alex Zhavoronkov said. “We’re using generative adversarial networks and reinforcement learning...a kind of AI imagination, used for applications where you need to generate new objects with a desired set of parameters.”
Insilico used knowledge of the virus’ structures gleaned from the 2003 SARS outbreak to begin its work early.
“De novo small molecule design is really desired, so we decided to go that route even though it takes longer and is much more risky,” Zhavoronkov said.
In its first iterations, the company initially saw mild results or even failed; but then as better crystal structures of targets were published, Insilico was able to generate compounds with promising effects.
“We can really show that now, in that context, AI can be a very powerful tool—to go from a new target to a novel small molecule that is not similar to anything else in the world,” he said. “COVID-19 actually presented us with an opportunity to test some of those techniques.”
“I must say that COVID has mobilized a lot of very smart people to go into this industry, so we are going to see a major wave of AI companies and technology giants get into this field.”