ERROR 1
ERROR 1
ERROR 2
ERROR 2
ERROR 2
ERROR 2
ERROR 2
Password and Confirm password must match.
If you have an ACS member number, please enter it here so we can link this account to your membership. (optional)
ERROR 2
ACS values your privacy. By submitting your information, you are gaining access to C&EN and subscribing to our weekly newsletter. We use the information you provide to make your reading experience better, and we will never sell your data to third party members.
Just over a month after the first reports of a novel coronavirus spreading in China, two independent groups this week reported that they have used artificial intelligence in different ways to find possible treatments for the virus. On Tuesday, researchers from the AI drug discovery company BenevolentAI and Imperial College London reported they have used AI software to find an already-approved drug that might limit the virus’s ability to infect people (Lancet 2020, DOI: 10.1016/S0140-6736(20)30304-4). On Thursday, the company Insilico Medicine announced its AI algorithms had designed six new molecules that could stop the virus from replicating in people’s bodies.
As news of the virus, called 2019-nCoV, spread since the end of December, scientists raced to learn more about it and find ways to fight it. Researchers in China published the first genomic sequence of the virus in mid-January, and by Jan. 29 scientists suggested that the virus might enter human cells by binding to a cell-surface molecule called angiotensin-converting enzyme 2, or ACE2 (bioRxiv 2020, DOI: 10.1101/2020.01.26.919985). The two AI drug discovery groups used this and other data to teach their software how to target the virus.
BenevolentAI’s algorithms connect molecular structure data to biomedical information about relevant receptors and diseases to find potential drug targets. The group adapted its search to the newly available information about 2019-nCoV and ACE2, and the software pointed to the enzyme adaptor-associated protein kinase 1 (AAK1) as a possible target for the disease. AAK1 regulates endocytosis, the process that brings material into cells, which also is a common mode of viral infection.
Next, the software and the researchers winnowed down 378 known AAK1 inhibitors to one possible drug, baricitinib, based on its affinity for the kinase and its toxicity. Baricitinib is approved to treat rheumatoid arthritis. The team rejected two cancer drugs that also might be effective because of their side effects and high predicted effective dosages. The researchers propose testing the drug in clinical trials against the virus.
But, in the paper, they also caution that their suggestion should not be taken as medical advice about 2019-nCoV treatment or prevention. They write that they published their research “to assist in the global response” to the virus.
Support nonprofit science journalism
C&EN has made this story and all of its coverage of the coronavirus epidemic freely available during the outbreak to keep the public informed. To support us:
Donate Join Subscribe
AI-based efforts like this could help to conserve drug researchers’ time and resources, says Mike Tarselli, scientific director of the Society for Laboratory Automation and Screening. “The use of AI to augment human capacity, to address a pressing public health concern using existing data without re-deploying a full team, should be a boon to researchers.” The BenevolentAI report is “more a testimony to good literature searching and curation” than “a triumph of artificial intelligence,” according to veteran drug-discovery researcher and blogger Derek Lowe. He says anyone could search through the extensive kinase literature to find good drug candidates, but he acknowledges that the researchers likely sped up their search with a well-organized database and software good at searching through it.
Timothy Cernak, a medicinal chemist at the University of Michigan, agrees with Lowe that most medicinal chemists could have probably identified this molecule as a potential drug using traditional database searches. He points out that as a class, kinase inhibitor drugs are known for targeting many different enzymes, meaning it wouldn’t be that hard to find one that acted against AAK1. He adds that the paper does not make clear how the researchers’ software decided to focus on that kinase, which could be a surprising result revealed by the AI.
Cernak says he’s both excited and scared by how quickly scientists have identified and published potential treatments for 2019-nCoV. He wonders what might happen if people worried about the virus take baricitinib without medical supervision; the US Food and Drug Administration describes the drug as a potent immunosuppressant with side effects that include serious or fatal infections. But he acknowledges that the researchers likely had few options except publishing what they found: “We’re in a global emergency, so we need answers.”
Insilico scientists took another approach to finding possible 2019-nCoV treatments. On Thursday, they released a paper describing a weeklong process using their AI algorithms to design new molecules that could limit 2019-nCoV’s ability to replicate in cells. Researchers studying the virus have already proposed doing so with approved antiviral drugs called 3C-like protease inhibitors. This protease helps the virus to communicate replication instructions to its host cell. The Insilico group argues that current knowledge of the virus protease suggests that those existing drugs would require unreasonably high dosages to be effective.
Instead, they set out to find possible novel drugs. The company’s AI algorithms evaluated the structure of the protease binding pocket and a representative inhibitor ligand for a virus that is nearly identical to 2019-nCoV, the coronavirus that caused severe acute respiratory syndrome (SARS) in 2003. The group programmed the algorithms to have a preference for novel structures with elements similar to known drugs. The paper describes six molecules that the group says are significantly different from known drugs but are predicted to be effective 3C-like protease inhibitors. The paper has been submitted to the bioRxiv preprint server but has not yet been accepted.
Five of the six do appear to be unique, according to Todd Wills, Managing Director at CAS, a division of the American Chemical Society, which publishes C&EN. The other has a central motif first described in 1910, he says, and which appears in the drug praziquantel.
The researchers say that they plan to start synthesizing and testing their molecules but that they are looking for outside partners who might want to help with those steps. Lowe is skeptical that these compounds will become medicines. “I would be surprised if anyone is willing to pony up enough money” to put Insilico’s molecules through a full drug development program, he says, considering that other, more proven drug design approaches are already underway for 2019-nCoV.
Insilico’s molecules appear to be made in part from reshuffled known drug motifs, says Ingo Hartung, director of medicinal chemistry at Merck KGaA, but that’s neither bad nor unexpected. Medicinal chemists often use such an approach. The promise of AI, he adds, is it will speed up the process of designing, testing, and making potential new drugs, which appears to be happening in this case.
What is needed now, Hartung says, is experimental data to evaluate the molecules: “Coronavirus patients don’t care how somebody came up with their drug but that the drug is efficacious and safe.”
On Feb. 11, 2020, the Coronavirus Study Group of the International Committee on Taxonomy of Viruses officially named the novel coronavirus "severe acute respiratory syndrome coronavirus 2" (SARS-CoV-2). The temporary name for the virus was 2019-nCoV.
This story was updated on Feb. 6, 2020, to include information about a second effort to find possible coronavirus drugs with artificial intelligence and to clarify a comment by Timothy Cernak.
Join the conversation
Contact the reporter
Submit a Letter to the Editor for publication
Engage with us on X