Merck & Co. has doubled down on its partnership with Variational AI, striking a deal worth up to $349 million to collaborate on small molecule candidates against two targets.
Variational disclosed a relationship with Merck early last year. At that time, the Big Pharma was assessing the ability of Variational’s Enki technology to generate small molecules that matched its target product profiles (TPPs). Enki uses generative artificial intelligence models to propose small molecules based on the TPP provided by the user.
Tuesday, Variational revealed that Merck’s early testing of the platform has led to a deal. Merck is paying an undisclosed upfront fee and committing to milestones that bring the potential total value of the deal up to $349 million.
In return, Merck has secured the chance to work with Variational to use Enki to design and optimize novel small molecule candidates against two undisclosed targets. Merck, which is responsible for picking the targets, will have the exclusive right to develop and commercialize compounds arising from the pact.
Variational CEO Handol Kim said in a statement that his team will apply its machine learning algorithms to Merck’s data sets to create generative models. The work will support attempts to drug two targets that Robert Garbaccio, Ph.D., head of discovery chemistry at Merck Research Laboratories, called “challenging.”
The deal is the latest in a series of Variational collaborations. Over the past year, the AI company has shared details of partnerships with Rakovina Therapeutics, ImmVue Therapeutics, Oncocross and Life Chemicals. Rakovina presented details of its use of Enki to generate brain-penetrating ATR inhibitors at the 2025 AACR Annual Meeting.
For Merck, the agreement adds another strand to its use of generative AI. The company shared details of another initiative in June, describing its work with McKinsey & Company and its AI arm, QuantumBlack, to create large language models for generating first drafts of clinical study reports. Merck said the models create drafts in three to four days, down from two to three weeks under the old approach.