
After nabbing two AI-generated molecules, AstraZeneca returns to BenevolentAI with new collaboration
Roughly three years ago, AstraZeneca teamed up with London’s BenevolentAI to bring new drugs into its portfolio using the biotech’s AI and machine learning capabilities. Now that the original deal has borne fruit, the two companies are re-upping their collaboration.
AstraZeneca and BenevolentAI will work together in two more disease areas: systemic lupus and heart failure, the companies announced Thursday. It’s a three-year partnership that comes after two candidates discovered by BenevolentAI for chronic kidney disease and idiopathic pulmonary fibrosis were nominated to AstraZeneca’s portfolio last year.
No financials were disclosed, but BenevolentAI CSO Anne Phelan told Endpoints News there would be an upfront payment, an equity investment and milestones.
The original partnership proved a significant milestone for AstraZeneca, as the CKD molecule nominated last January was the company’s first generated by AI. At the time, AstraZeneca identified and validated a new way to attack CKD thanks to the biotech’s platform AI tech, and began developing compounds centered on this target.
The IPF compound followed up just last month, and now marks the “perfect time” to extend the working relationship, Phelan said.
Lupus and heart failure “fall within [AstraZeneca’s] priority therapeutic areas,” Phelan said. “They’re complex diseases, difficult spaces to work in. It’s a tricky problem that lends itself well to AI and our machine learning capabilities.”
At the heart of BenevolentAI’s technology is what Phelan describes as a “knowledge graph,” which essentially looks like a “hairball” network of thousands of data points, or nodes, interconnected by millions of lines or relationships. Such relationships are first identified by the machine learning algorithms and can connect many different and unrelated diseases.

BenevolentAI is able to differentiate itself from the buzzy and growing field of AI drug development because it is driven by a hypothesis-first approach rather than focusing on phenotypic screening, Phelan added. Whereas competitors’ AI might be making new molecules and trying to reverse-engineer the process, BenevolentAI starts from scratch.
“It’s not a right and wrong situation … it’s a really multidimensional picture of human biology across the commonalities of human disease,” Phelan said. “For example, depending on the cell type that’s affected, the graph could end up connecting to Alzheimer’s or diabetes, where underlying physiology could be really quite similar.”
For AstraZeneca, lupus recently turned into a potential moneymaker, as the Big Pharma whisked a new drug past the FDA finish line last August in Saphnelo. It’s the first new lupus approval in at least the last 10 years, AstraZeneca said at the time, and reduces overall disease activity across organ systems and oral corticosteroid use compared to placebo.
Heart failure, meanwhile, is an area where AstraZeneca is looking to stake out a blockbuster with Farxiga. Back in 2020, the drug was approved to reduce the risk of cardiovascular death or hospitalization in heart failure patients with a reduced ejection fraction. It also picked up an approval in CKD last April.