Industrial biotech · Enzyme design

Ligo Biosciences

A San Francisco startup building generative diffusion models to design enzymes from scratch. Its first public move was not a product, it was an open source rebuild of Google DeepMind's AlphaFold3.

A pub in Oxford, 2021

The story Ligo tells about itself starts before the company did. In 2021, weeks after AlphaFold2 launched, three Oxford students met in a pub and started bouncing around ideas about where artificial intelligence and biology might collide in a way that mattered. Edward Harris was studying medicine after transferring from computer science at Princeton. Emily Egerton-Warburton was a biochemist getting her hands wet in synthetic biology labs. Arda Göreci was studying cell and systems biology and had, in his own words, become obsessed with deep learning for biomolecular design the moment the original AlphaFold paper came out.

Therapeutics were the obvious place to point that obsession, and the one they rejected. “We didn’t want to wait ten years for regulatory approval,” Harris told a reporter later. “We wanted to build something people could use now.” The something became enzymes.

The problem: catalysts that take years to build

The company frames the target in blunt terms: a $6 trillion chemical industry that produces 20% of industrial greenhouse gases and consumes 15% of global energy. Enzymes, nature’s catalysts, already make a narrow slice of that industry cleaner. Pharmaceutical manufacturers use them for a limited set of drugs. But building a new enzyme still mostly runs through directed evolution, the technique that won its inventor a Nobel Prize but works by mutating DNA, testing thousands of variants, and slowly refining results over months and, often, millions of dollars.

Ligo's pitch is that no existing model actually understands catalysis, so today's tools can only nudge an enzyme that already exists rather than design one from nothing. The company says its models are trained to generate structures capable of catalysing a reaction directly from a transition state model, in days rather than years. That claim has not been independently tested in a published, peer reviewed study as of this writing.

Three founders, three disciplines

Harris, the CEO, is a second time founder. Before Ligo he bootstrapped a company called Abas2Go in the food markets of Guadalajara, Mexico, taking it, by the company's account, to $1 million in revenue. Egerton-Warburton, the CSO, is described on her own Oxford lab page as a Part II Biochemistry student at St Peter's College who joined the Higgins Lab in 2024 to work on structure guided vaccine design, after earlier stints on cyanobacteria genetic engineering at CyanoCapture and a pneumococcal vaccine candidate at Inventprise; that page also notes she has been working on de novo enzyme design using generative diffusion models, which lines up with Ligo's technical pitch. Göreci, the CTO, studied cell and systems biology at Oxford and was named a Google Cloud Research Innovator for computational biology work.

The AlphaFold3 stand

Ligo's public breakout did not come from a product launch. In 2024, Google DeepMind published the AlphaFold3 paper in Nature without releasing the source code. Ligo rebuilt the model from the paper and released it publicly on GitHub as an open source project, along with training code, under an Apache 2.0 license. In doing so the team said it found places where the published supplementary information appeared inconsistent with the rest of the deep learning literature, including a possible error in how the MSA module communicates with the pair representation, a loss scaling factor that did not match the properties described by the method it cited, and a DiT block design that omitted residual connections other implementations use. The repository lays these out in detail and invites the community to weigh in.

The release credits close technical partnerships: the OpenFold project supplied reusable modules and data pipelines, and Ligo says it partnered with Adaptyv Bio, whose engineers Liza Kozlova and Igor Krawczuk are helping build a version of Adaptyv's ProteinFlow data pipeline with full ligand and nucleic acid support. Basecamp Research is named as a second data partner, contributing what Ligo describes as sequence diversity roughly a thousand times beyond public databases to improve a forthcoming open source model.

“It showed we had the technical capabilities.” the founders said of the release. The GitHub repository states plainly that the work is an active research project in its early phase, not yet production ready: it currently predicts single chain protein structures only, the same scope as the earlier AlphaFold2, with ligand, multimer and nucleic acid prediction still to come.

From a pub idea to Silicon Valley

The AlphaFold3 release, and the attention it drew on social media, is the version of events Ligo and outside coverage both point to as the reason the company applied to Y Combinator. Harris recalls being surprised to land an interview at all given the accelerator's roughly 2% acceptance rate, and describes the interview itself, led by a YC group partner, as an unexpectedly intense test of whether the founders were serious about dropping out and committing full time. A call came at midnight offering them a spot.

Ligo says it closed its initial funding round in six hours, at a valuation the founders describe as multiple times higher than offers they had received in the UK. As of the most recent independent reporting found for this profile, the exact size of that round had not been disclosed publicly.

What is proven, and what is still claimed

Read plainly: the clearest, most verifiable thing Ligo Biosciences has done in public is release a working, documented reimplementation of a major DeepMind model, with named technical partners and specific, checkable claims about where the original paper's description is unclear. Its core commercial claim, that its models can design working enzymes from transition states in days, has not yet been tested against a published dataset that anyone outside the company can see.

What to watch

• Whether Ligo publishes a preprint or benchmark showing its enzyme design models producing enzymes with measured activity, not just structure predictions.
• Ligand, multimer and nucleic acid support for the open source AlphaFold3 release, which the company has said is still in training.
• Named paying customers in pharmaceuticals, agriculture or consumer chemicals, the industries the company has said it wants introductions to.
• Public disclosure of the seed round size, which as of this profile has not been released.

In their words

Related companies

• C16 Biosciences: Also in Industrial Bio. A New York company that brews a palm oil substitute in yeast rather than growing it on cleared rainforest
• Solugen: Also in Industrial Bio. A Houston company that turns corn sugar into industrial chemicals using an enzyme its co-founder found in pancreatic cancer cells
• Rebellyous Foods: Also in Industrial Bio. A Seattle company that builds the machines that make plant-based chicken, and sells the chicken mostly into school cafeterias
• Kern Systems: Also in Industrial Bio. A Boston company spun out of George Church’s Harvard lab to write digital data into DNA with enzymes instead of chemistry

Sources

1. CompanyLigo Biosciences homepageligo.bio · accessed Sep 25, 2026
2. CompanyLigo Biosciences research blogresearch.ligo.bio · accessed Sep 25, 2026
3. CompanyLigo Biosciences profile and launch tl;drY Combinator · accessed Sep 25, 2026
4. CompanyLaunch YC: Ligo Biosciences, generative enzyme design for the chemical industryY Combinator · 2026
5. Public recordLigo-Biosciences/AlphaFold3 repository and READMEGitHub · accessed Sep 25, 2026
6. IndependentThe Startup That Took on DeepMind and Won the Attention of Silicon ValleySustainable Times, Daisy Moll · May 7, 2025
7. IndependentArda Goreci, Edward Harris, Ligo Biosciences (podcast episode)Sustainable Times · May 7, 2025
8. IndependentLigo Biosciences: Advancing Green Enzyme TechnologyHiretop, Daryna Falko · Aug 14, 2024
9. IndependentLigo Biosciences Launches: Generative Enzyme Design for the Chemical IndustryFondo · Feb 9, 2026
10. IndependentEmily Egerton-Warburton, lab member pageHiggins Lab, University of Oxford · accessed Sep 25, 2026
11. IndependentLigo Biosciences company profileTracxn · accessed Sep 25, 2026
12. IndependentLigo Biosciences company profilePitchBook · accessed Sep 25, 2026
13. Public recordEDGAR company search, companies matching "Ligo"SEC · accessed Sep 25, 2026