The Wednesday Pharma R&D
Roche will use Chai Discovery's AI to design protein drugs across several programs
Chai Discovery, a San Francisco company that designs protein-based medicines with AI, signed a collaboration with Roche. It is the latest of the pharma deals Chai has announced this year.
By Matteo Favilli, 2026-11-04
In short
Roche will use Chai Discovery's AI models in a number of its drug discovery programs. The models design new protein-based drug candidates, such as antibodies, on a computer. The companies did not disclose targets, money or timelines, and no results from the collaboration have been published.14
Why it matters
Roche joins a list of large drugmakers working with Chai this year. In October, Chai said it would partner with GSK, and in June Pfizer said it would use Chai's platform to speed up its drug discovery research.3
Protein-based medicines, such as antibodies and engineered proteins, are a large part of modern drug development. Finding them usually takes repeated cycles of experimental screening, molecular engineering and lab validation.4
How it works
Chai builds what it calls a computer-aided design suite for protein-based therapeutics. Its models reason about biological structure and function and generate new molecular designs from scratch, with the aim of reaching targets that traditional discovery methods have struggled to reach.1
Chai-3 is the company's latest model. According to Chai, it achieves high target success rates and binding affinities, including to challenging targets. Binding affinity is how tightly a molecule attaches to its target.1
Under the collaboration, Roche will use Chai's models and platform in a number of drug discovery programs across its therapeutic areas.1
What they found
The Roche announcement contains no data. Neither company disclosed the targets, disease areas, number of programs, financial terms or timelines, and Chai has not published performance figures or independent benchmarks for Chai-3 in this collaboration.4
The closest public evidence comes from GSK. Before its own deal, GSK tested Chai's models on its targets: the models generated designs zero-shot, meaning without training specific to each target, and GSK's internal wet-lab testing found binders to all tested targets.2
Christopher Austin, who leads R&D technologies at GSK, said the company saw designs with strong binding affinities, and that Chai's technology complements GSK's internal models, which it keeps retraining on its own wet-lab data.2
Matteo's take
In practice, Roche is adding an outside design tool to its early discovery work. The model proposes protein candidates on a computer, and Roche's labs still have to make and test them. GSK took the same path in a visible way: it ran Chai's designs through its own wet lab first, and expanded the work after the results came back.
The caveats are the usual ones for an early deal. The announcement does not disclose targets, financial terms or timelines, and the claims about Chai-3 come from Chai itself. Computer predictions still need experiments to show activity, specificity and stability before a design can become a drug candidate.
The detail I would keep in mind is how GSK described its setup: Chai's models sit next to internal models that GSK retrains on its own lab data. A company with Roche's range of programs can do the same, and test Chai's designs against its own models on the same targets.
Sources
Drafted with AI from the sources below.
- Chai Discovery Announces Collaboration with Roche in AI-Driven Drug Discovery, Chai Discovery
- Chai Discovery Announces Collaboration with GSK Following Successful Wet-Lab Evaluation, Chai Discovery
- Chai Discovery, Roche to Collaborate on AI Drug Discovery, Contract Pharma
- Chai Adds Roche to Its Expanding AI Drug Discovery Network, Pharmacally
All of AI Bits