AI drug race moves from finding candidates to proving they work

AI is moving beyond screening vast pools of compounds. It identifies promising drug candidates and draws pharmaceutical companies into closer partnerships with AI firms.

Inside a lab at the Advanced Radiation Technology Institute in Jeongeup, North Jeolla, Novorex researchers were busy testing AI-selected drug candidates on Sept. 10.

To see whether AI-predicted compounds actually bind to proteins and produce the intended effect, researchers used E. coli and other organisms to make disease-associated proteins. For example, a protein or gene that fuels tumor growth can become the target a drug is designed to hit.

“Even if AI identifies a highly promising candidate, we still need high-purity proteins to determine whether the drug actually binds to its intended target,” Novorex CEO Son Woo-sung said.

AI has made it faster to hunt for promising drug candidates. Now the race is moving to the next stage: validating them in the lab and developing them into actual medicines.

Pharmaceutical companies are stepping up partnerships with AI drug developers. The potential value of related deals reached nearly 63 trillion won ($46.4 billion) in the first half of this year alone.

Novorex found its Parkinson’s disease drug candidate NRX-NGT002 in about 18 months. Using AI, it synthesized only 186 compounds along the way. A conventional search would have required roughly 2,000.

The candidate was selected for a Michael J. Fox Foundation support program alongside Biogen and GSK. Novorex is targeting global clinical trials by year-end.

“We placed a high value on Novorex’s ability to conduct biophysical validation and generate drug assets that can actually be developed,” said an SK Biopharmaceuticals official. The company selected Novorex as a collaboration partner in April.

AI biotech company Portrai signed a deal worth up to 125.9 billion won with Celltrion to discover new drug targets. Its validation technology helps determine which targets are most likely to respond to a drug.

Portrai first uses AI to scan tumor tissue maps for promising drug targets. It then checks those targets against patient data to see whether they hold up in humans.

“Early on, the focus was on the AI algorithm itself,” Portrai CEO Lee Dae-seung said. “Now, as the industry learned that even good algorithms can be wrong, the emphasis has shifted toward validation.”

AI is moving beyond the hunt for promising compounds. As discovery, validation, clinical trials and drug development become more tightly linked, global pharmaceutical companies are ramping up partnerships with AI drug developers.

AI drug research and development (R&D) partnerships totaled 70 in the first half of the year, according to biotech data provider DealForma. Their potential deal value reached $45.9 billion — topping the $43.4 billion recorded for the whole of last year.

SK Biopharmaceuticals signed a deal worth up to $2.57 billion with Insilico Medicine, an AI drug developer that identifies promising compounds and advances them toward clinical trials.

Seoul-based Galux is also working with AstraZeneca to validate AI-designed proteins experimentally. Proteina, another Korean biotech company, is collaborating with Samsung Bioepis and U.S.-based Twist Bioscience.

The push to strengthen validation is taking different forms. Novo Nordisk teamed up with Anthropic last Wednesday to bring the AI company’s Claude Science model into its drug R&D.

Anthropic, meanwhile, is building a life sciences wet lab in the San Francisco Bay Area, where AI predictions can be paired with physical experiments.

Big potential deal values mask a more cautious approach to upfront spending. The median upfront payment per AI drug R&D deal fell to $15 million in the second quarter this year from $78 million in the first, according to DealForma.

“Deal values will rise as AI improves the odds of successful development and cuts risk,” Lee of Portrai said.

BY CHOI EUN-KYUNG [[email protected]]