Foundation Model for Chemical Manufacturing
Yoneda Labs · Coding
Yoneda Labs' Foundation Model for Chemical Manufacturing predicts the reaction conditions that lift yield, learning from real bench data instead of guesswork.
About Foundation Model for Chemical Manufacturing
What Is Foundation Model for Chemical Manufacturing
Yoneda Labs builds a foundation model for chemical manufacturing, and the name is the pitch: a model trained on real lab experiments that tells a chemist which temperature, concentration, catalyst, ligand, base, and solvent will actually make the molecule they need.
The company started in 2023 in San Francisco. All three founders came out of the University of Cambridge, and the team raised a $4M seed led by Khosla Ventures in April 2024. Prof. Scott E. Denmark advises the group.
The problem it attacks is old and expensive. Chemists pull conditions from literature that almost never lists the exact recipe, then burn weeks changing one variable at a time. Yoneda's answer is data it generates itself: AI for chemistry built from thousands of high-throughput experiments feeding a planner that decides which experiment to run next.
Limits deserve equal billing. Coverage leans on organic transformations the team has actually run, so an unusual reaction class can fall back to ordinary screening. Pricing isn't published, there's no self-serve signup, and the software enters a company through demos and pilots. Small academic groups will find it out of reach.
Getting Started
- Book a demo through the Yoneda Labs site and bring a reaction you already struggle with. The team scopes whether software alone or a lab campaign fits.
- Load your existing experimental records, or start from a blank substrate if you have no data yet.
- Set the variables your equipment can actually change, such as temperature, solvent, ligand, base, and catalyst loading.
- Review the recommended batch of experiments, then run it at your bench or hand it to Yoneda's high-throughput lab.
- Log the yields back in. The model updates and proposes a smaller second round.
Product Information
A quick look at Foundation Model for Chemical Manufacturing's pricing, supported platforms, and performance.
Best for
The users, tasks, and scenarios where this tool fits best.
Users
- Medicinal chemists
- Process chemistry teams
- CRO and CDMO teams
Tasks
- Rescuing a low-yielding cross-coupling
- Choosing between candidate routes
- Building a library on general conditions
- Moving conditions toward manufacturing
Scenarios
- A weekly optimization cycle with no automation in house
- Early route selection on a new chemical series
- A team without a dedicated high-throughput group
Key features
Reaction Yield Prediction
Hand the model a substrate pair and it ranks condition sets by expected yield before anything touches glassware. The ranking comes from tens of thousands of in-house experiments, mostly on transition metal catalysis. In a 2025 collaboration with the CRO Symeres, the recommended conditions moved a cross-coupling from roughly 30% to above 90% in situ yield.
Bayesian Experiment Planning
Every round of experiments is picked to teach the model the most. Bayesian optimization spends a little effort on unexplored conditions and most of it on promising ones, so a campaign often lands on workable numbers in two or three rounds. Symeres reported four diverse conditions above 90% yield, found faster than their usual one-factor-at-a-time or DoE sequence.
Software That Learns From Your Bench
The Yoneda app, the company's reaction optimization software, doesn't need a full historical dataset to start. It absorbs the results your chemists already generate, so the model sharpens as the project runs. A group running twenty reactions a week ends up with a private model of its own chemistry.
Offline Deployment for Sensitive Data
Pharma companies won't push structures into a cloud they don't control. The app runs fully offline on a chemist's PC, and the custom algorithms stay local. That design choice is often what gets a pilot past an information security review.
High-Throughput Lab Services
Yoneda runs its own experimentation lab: glovebox chemistry under inert atmosphere, automated solid and liquid dispensing, plus LC-MS, NMR, and prep-HPLC in house. Teams that would rather not build automation can hand over a campaign and get back purified compounds or a conditions table. The lab doubles as the data engine for the model.
Route Scouting and Library Work
Retrosynthesis tools return five plausible disconnections and no ranking you can trust. Yoneda runs candidate routes head to head with real material and reports which one survives contact with the bench. The same team covers library enablement, where a representative substrate scope turns into a handful of general conditions, and full library synthesis with purification.
Pros and cons
Pros
- Recommendations arrive as a short list of experiments, not a 500-well plate to interpret.
- Yield movement shows up in weeks: the Symeres cross-coupling went from about 30% to above 90%.
- Works on data you already own, and the model can stay on your machine.
- The lab side lets you outsource screening instead of buying robotic hardware.
- Reachable for mid-size process groups that can't staff a dedicated high-throughput team.
Cons
- Reaction coverage is narrow next to the 10M+ compounds synthesized each year, so unusual chemistry still needs manual screening.
- No public pricing and no self-serve trial means every evaluation starts with a sales call.
- No documented public API, which blocks teams that want the model inside their own electronic lab notebook.
- Predictions come from small-scale plates, so mixing, heat transfer, and other scale-up problems still land on your engineers.
Frequently asked questions
It predicts reaction outcomes. For a given substrate pair it ranks combinations of temperature, concentration, catalyst, ligand, base, and solvent by expected yield, and it picks the next experiments that will teach it the most.
Related content
Explore related tools, skills, and articles for Foundation Model for Chemical Manufacturing.
Foundation Model for Chemical Manufacturing Alternatives
Base Code
Base44 (Wix) · Coding · ProductivityBase Code is a shared cloud development environment from Base44. Teammates describe changes to an existing GitHub codebase in chat, preview them live and send them back as pull requests.
Google EmbeddingGemma 2
Google DeepMind · Writing · Image · CodingGoogle EmbeddingGemma 2 is a 740M-parameter open model that turns text, code, images, audio, and video frames into vectors in one 768-dimension space, running on phones and laptops.
GitGlow
GitGlow · Coding · ProductivityGitGlow is a local-first desktop Git client that lets independent developers review their own code and their AI agents' changes in one place before shipping.