Home  >  News

AI invents new ‘recipes’ for potential COVID-19 drugs

Aug 09, 2020

Science's COVID-19 reporting is supported by the Pulitzer Center and the Heising-Simons Foundation.

As scientists uncover drugs that can treat coronavirus infections, demand will almost certainly outstrip supplies—as is already happening with the antiviral remdesivir. To prevent shortages, researchers have come up with a new way to design synthetic routes to drugs now being tested in some COVID-19 clinical trials, using artificial intelligence (AI) software. The AI-planned new recipes—for 11 medicines so far—could help manufacturers produce medications whose syntheses are tightly held trade secrets. And because the new methods use cheap, readily available starting materials, licensed drug suppliers could quickly ramp up production of any promising therapies.

“If you are going to supply a drug to the world, your starting materials have to be cheap and as available as sugar,” says Danielle Schultz, a chemist at Merck. The new method, posted as a preprint this week, “is really solid,” she says. “I am impressed by the speed at which [the researchers] were able to find new solutions for making existing drugs.”

Patents give pharmaceutical companies the right to be the sole supplier of a new drug in a given country, usually for 20 years. Once a drug goes off patent, other companies can produce and sell it as a generic. The method to make the drug is often secret to discourage competition even after patents expire. But COVID-19 has changed all that, Schultz says. “We are at a time when it’s all hands on deck.”

Only two medicines—remdesivir and dexamethasone—are currently proven to fight COVID-19. That has led to supply shortages for both. On 4 August, attorneys general from 34 U.S. states wrote federal officials, calling remdesivir supplies “dangerously limited,” and urging states be given “march-in rights” to violate owner Gilead Sciences’ patents. Such rights would allow states to work with third-party manufacturers to make additional supplies of the drug.

To prevent future supply crunches, University of Michigan chemist Timothy Cernak and colleagues turned to a commercial drug synthesis AI program called Synthia. The software can help pharmaceutical manufacturers find the most efficient and cost-effective strategy for synthesizing medicines, most of which are fairly complex molecules that can be built in myriad ways—much as an artist can apply brush strokes in infinite combinations to paint the same landscape. “It’s more options than the human mind can comprehend,” Cernak says.

Cernak and his colleagues scoured the research and patent literature for ways to synthesize 12 medications now being tested as COVID-19 therapies, including remdesivir. They then programmed Synthia to search for new synthetic solutions. They limited their search to options that used cheap, abundant starting materials, didn’t require expensive catalysts or equipment, and could produce kilogram-scale amounts of drug.

In the end, the software found novel solutions for making 11 out of the 12 compounds, including generic antivirals umifenovir and favipiravir, the researchers report this week in a non-peer-reviewed preprint on ChemRxiv. The AI program came up with four different ways to synthesize umifenovir, for example, in one case with cheaper starting materials than those currently in use. “For the same amount of money [or less], we can make these drugs from different starting materials,” Cernak says. The one miss was remdesivir: The software was unable to come up with a solution for making it other than the way than Gilead does, he says. Cernak says he and his team filed patents on all of their new synthetic routes. But their goal isn’t to make a profit. Instead, they want to license their manufacturing approaches to one or more pharmaceutical companies to ensure adequate supplies and low prices.

Now, he adds, they wait and see whether any of the drugs prove effective in clinical trials.

Posted in: HealthTechnologyCoronavirus


Previous: None

Next: Glycan recognition in globally dominant human rotaviruses