Late one January afternoon, British pharmacologist Peter Richardson ran out of his home office and told his wife, “Got it!” She asked what he was talking about and offered a cup of tea. Richardson explained that he had identified a drug that might help people infected with a new virus spreading in China.

Richardson’s dash was prompted by a finding from artificial intelligence software developed by his employer, BenevolentAI, a London startup where he is vice president of pharmacology. The company has created a kind of search engine on steroids that combines drug industry data with nuggets gleaned from scientific research papers. Using the software, Richardson had identified a rheumatoid arthritis drug that might dampen some of the most severe effects of the new virus, an illness now known as Covid-19.

Read all of our coronavirus coverage here.

The virus, and that idea, have advanced rapidly in the weeks since. In February, Richardson and others at BenevolentAI published two research papers laying out their hypothesis and supporting evidence. They caught the attention of Eli Lilly, which markets the arthritis drug, known as baricitinib, under the brand name Olumiant.

This week, Lilly announced it is working with the US National Institute of Allergy and Infectious Diseases on a large clinical trial of the drug in hospitalized Covid-19 patients. Patrik Jonsson, president of Lilly’s biomedicines division, says his group hadn’t previously thought of baricitinib as an infectious disease treatment. “I think Covid-19 in many ways will change the way we’re getting work done,” he says.

The clinical trial should begin in the US this month and could expand to include patients in Europe and Asia. Results are expected as soon as late June. Jonsson says it usually takes years to design, organize, and launch a trial.

The rapid progression from initial idea to clinical trial shows how widely researchers and drug companies are looking as they scramble to stem the coronavirus pandemic. “I can’t guarantee that baricitinib will work out OK, but there’s huge unmet need,” Jonsson says. “We don’t know how to treat these patients.”

The tale also highlights the potential for computing and artificial intelligence to help that effort. Since the 1950s, the time and cost of developing new drugs have increased exponentially, partly because of higher safety standards. Some investors and pharmaceutical companies believe computing power and algorithms can shorten the development cycle in some cases.

Lilly and fellow drug giant Pfizer have partnerships with Silicon Valley startup Atomwise, which uses machine-learning technology to find novel compounds that target particular biological molecules. Last year Atomwise helped Stanford researchers find a way to target an enzyme that they had discovered accumulates in the cells of patients with Parkinson’s Disease.

BenevolentAI has similar technology and its own Big Pharma partnerships, with Novartis on cancer and AstraZeneca on kidney disease.

When Richardson and others at BenevolentAI decided to take on the new coronavirus, they hoped to find an existing drug that could be repurposed, in order to reduce safety and regulatory hurdles. But they didn’t know much about the enemy. The virus was–and in many ways still is–too new to have been fully characterized.

Richardson began his explorations in BenevolentAI’s system by seeking ways to interfere with the mechanism by which related and better-known coronaviruses, such as SARS, invade a person’s cells. BenevolentAI’s software can offer interactive visualizations of the connections among diseases, symptoms, and biological processes, sourced from databases and machine-learning algorithms that process text in scientific papers. The colorful web of proteins and genes that Richardson conjured up presented some promising targets.

“Hanging off the bottom of the graph was a pale blue section that leapt out of the page,” Richardson says. It was a clump of genes that regulate the cellular machinery a coronavirus exploits to enter and infect a cell. Gumming up that machinery by targeting those genes with a drug might slow the virus.

Richardson’s “Got it!” moment came after he searched for approved drugs that would specifically target two of the most crucial genes in that clump and might be effective with a small dose. “One just floated right to the top,” Richardson says: baricitinib.

BenevolentAI’s hypothesis was published in a letter to the prominent medical journal The Lancet early in February. A more detailed follow-up cited further results from the company’s knowledge base suggesting that the anti-inflammatory mechanisms that make baricitinib effective against rheumatoid arthritis might help quell the out-of-control immune response dubbed a cytokine storm that can damage the lungs and other organs of people with severe cases of Covid-19. Baricitinib is effective for rheumatoid arthritis because it inhibits a protein involved in the over-the-top immune response that causes the disease’s characteristic joint pain. The same protein is involved in cytokine storms like those seen in Covid-19.

Keep Reading

illustration of a head

The latest on artificial intelligence, from machine learning to computer vision and more

Jonsson says the unexpected suggestion from BenevolentAI prompted Lilly to examine its data on baricitinib. The company also talked with outside researchers who did new lab tests on the drug. Lilly’s experts concluded that BenevolentAI’s hypotheses had merit, particularly the notion that the drug might dampen Covid-19’s dangerous cytokine storms. The company found corroborating evidence in results from early tests of baricitinib in severe Covid-19 cases by doctors in Italy inspired by BenevolentAI’s work. That’s when it opened talks with the government’s infectious disease institute about a trial to test the drug’s effect on Covid-19 patients.