Variational AI CEO Handol Kim says the deal is important for Canadian productivity and competitiveness.Jennifer Gauthier/The Globe and Mail
Variational AI Inc., a Vancouver startup that uses generative artificial intelligence to discover therapeutic drugs, has signed a two-year development deal with Merck & Co. Inc. MRK-N that could be worth up to US$349-million.
Variational will use its Enki platform, trained on Merck data, to generate promising molecules for undisclosed “challenging therapeutic targets” selected by the American drug giant, said Robert Garbaccio, vice-president and head of discovery chemistry with Merck Research Laboratories in a release Tuesday.
Merck will have the exclusive right to develop and take to market any compounds arising from the collaboration, which stems from a partnership the companies struck last year. Merck also invested in a US$5.5-million venture capital financing of Variational this year, which the startup will spend on increasing its compute capacity to power its algorithms.
The parties didn’t reveal the deal terms, but Variational will get an upfront payment plus additional sums for reaching certain milestones. Variational AI chief executive officer Handol Kim said his 18-person company wouldn’t receive royalties from sales of any resulting drugs that make it to market.
The deal “is a validation for both AI and drug discovery and specifically for our model and approach,” Mr. Kim said in an interview. “It’s important for Canadian productivity and competitiveness. We’re a country of biotechs but have no big pharma companies. AI is one of the best ways we can be globally competitive within the biotech and life sciences industry.” Mr. Kim added Variational AI is working with several other partners on similar deals.
There has been a lot of hope – and hype – around the ability of AI to revolutionize drug discovery. So far no AI-generated molecules have been approved by regulators for commercial sale.
The most advanced AI drug developer is Boston-based Insilico Medicine Inc., whose drug rentoserib, which targets a chronic lung disease, was shown to be safe and well tolerated in human trials this year. Insilico now plans to conduct late-stage, expanded human efficacy trials, a key step toward gaining regulatory approval. Insilico, which has a research and development centre in Montreal, has ambitious plans to develop 22 molecules and has received U.S. Food and Drug Administration approval to begin clinical studies on 10 of them.
Insilico is also one of several AI drug discovery companies, including Alphabet Inc. GOOGL-Q subsidiary Isomorphic Labs, that have signed licensing deals with big pharma companies.
Two British AI startups, BenevolentAI SA and Exscientia PLC, have encountered setbacks. An eczema drug developed by BenevolentAI failed to reduce itch and inflammation in trials in 2023, while Exscientia shuttered trials of its cancer treatment drug and cut jobs before the company was purchased by Recursion Inc. RXRX-Q, which itself has subsequently ended some discovery programs.
Drug discovery is notoriously hard; 90 per cent of candidates fail clinical trials. AI cannot change the expensive and time-consuming animal and human trials crucial to proving a compound is effective and safe. And AI drug companies don’t have the same ready access to abundant information to draw from as large language models such as OpenAI’s GPT products, which train on massive quantities of text from the internet. Data that does exist for AI drug companies can be of poor quality and needs much curating to be useful, and experts in AI are not typically steeped in pharmaceutical industry knowledge.
Variational was founded by AI developers from D-Wave Quantum Inc. after the Burnaby, B.C.-based company shut their unit in 2019, which was led by Mr. Kim, in 2019. The team had no background in drug discovery.
Variational’s Enki platform works like a large language model, but instead of producing text answers or pictures, it generates novel small molecules. A user can input the desired properties of a molecule and Enki will produce structures that meet the criteria. Those are investigated further to see if they have the potential to become drugs. It typically costs Variational hundreds of thousands of dollars and takes just weeks to generate potential molecules, rather than millions of dollars and years in typical drug discovery work, though that doesn’t guarantee their success in trials.
Mr. Kim said Variational will continue to focus on AI and leave the laboratory work to third-party contract research organizations.
Editor’s note: A previous version of this article incorrectly stated that Merck previously signed partnership deals with British AI startups BenevolentAI SA and Exscientia PLC. Those deals were made with the unaffiliated company Merck KGaA.