Loading
Old version — revision 1
This is a fixed snapshot of Isomorphic Labs, saved by Import as part of the initial corpus import. It is not edited and it is not updated; the article may have changed since.
Edit summary: Initial import of content/isomorphic-labs.md — the filesystem corpus, unchanged. Not an edit.
Alphabet's London-based AI drug-design company, founded in 2021 by Demis Hassabis to build on AlphaFold; it co-developed AlphaFold 3 and has no drug in human trials as of October 2026.
Isomorphic Labs is an Alphabet-owned company in London that designs drug candidates with machine-learning models, founded in 2021 by Demis Hassabis to carry DeepMind's protein structure work into pharmaceutical research. It co-developed AlphaFold 3 with Google DeepMind, has research collaborations with Eli Lilly, Novartis and Johnson & Johnson, and has raised $2.7 billion in external funding rounds. As of October 2026 no molecule it designed has been given to a person, so the clinical value of its approach is still untested. It is one of the best-funded tests of whether structure prediction can be turned into medicines.
Isomorphic presents drug discovery as an engineering problem. Its models predict how a candidate molecule binds its target and what properties it will have, so that fewer compounds need to be made and tested in the laboratory. It is a vertically integrated form of AI drug discovery: the company builds its own models, generates its own data and runs its own medicinal chemistry, and it says it will take candidates into clinical trials itself rather than only licensing them out. Its stated mission is to "solve all disease", an aspiration with no timeline attached.
The company works on two tracks. Partnered programmes with large drug companies bring in money and targets chosen by the partner. Wholly owned internal programmes, which the company describes as focused on oncology and immunology, are where any clinical candidate of its own would come from. It says its engine now designs antibodies, peptides and molecular glues as well as small molecules.
Hassabis runs Isomorphic while remaining chief executive of Google DeepMind. Max Jaderberg, a founding member who had led the company's AI research as chief AI officer, became president on 1 January 2026, replacing Colin Murdoch, who moved to an advisory role. By late 2025 the company reported more than 300 staff across London, Lausanne and Cambridge, Massachusetts.1 Its scientific advisory board includes Jennifer Doudna and Venki Ramakrishnan.
Alphabet announced Isomorphic in November 2021 as a separate company from DeepMind with its own resources. It was built on AlphaFold 2, the DeepMind system that in 2020 had predicted single-chain protein structures to near-experimental accuracy. The company says many of its staff contributed to AlphaFold.
The first commercial deals came in January 2024. Eli Lilly agreed to pay $45 million upfront and up to $1.7 billion in milestones, and Novartis $37.5 million upfront and up to $1.2 billion, plus royalties in both cases. Both deals covered small-molecule programmes against undisclosed targets.2 In February 2025 Novartis added up to three more programmes to its original three. A multi-target collaboration with Johnson & Johnson followed in January 2026, with no financial terms disclosed: Isomorphic does the computational design, and Johnson & Johnson runs the assays and takes programmes forward.
The company's best-known output is AlphaFold 3, published in Nature in May 2024 by Google DeepMind and Isomorphic authors. It replaced AlphaFold 2's structure module with a diffusion-based generator and extended prediction from proteins to complexes with DNA, RNA, small molecules, ions and modified residues. The paper reported much better accuracy on protein–ligand complexes than standard docking tools, which is the case that matters most for drug design.3
The release was disputed. The paper appeared without code, and outside access was limited to a web server that restricted which ligands could be modelled. Scientists objected that this undermined reproducibility. DeepMind said the restrictions balanced research access against the protection of commercial interests, and Isomorphic applies the model in its own drug discovery. In November 2024 the code was released for non-commercial use, with trained weights available on request to researchers at academic institutions only.4 The weights' terms still bar commercial use, and commercial users are directed to a paid Google Cloud service. The Nobel Prize in Chemistry that Hassabis and John Jumper shared that October, with David Baker, honoured AlphaFold 2 and named both men as Google DeepMind researchers.5
In February 2026 the company published a technical report on what it calls the Isomorphic Labs Drug Design Engine, or IsoDDE. The report claims more than twice AlphaFold 3's accuracy on the hardest protein–ligand cases in a benchmark built to test unfamiliar pockets and ligands. It also claims binding-affinity predictions better than physics-based free-energy methods, and the ability to find binding pockets on a protein from its sequence alone.6 The accompanying announcement concedes a problem with AlphaFold 3: its accuracy fell on structures unlike its training data, which is where many new drug targets sit. In a September 2026 essay Jaderberg described design agents that search chemical space on a computer for two to four days, after which a handful of compounds are synthesised and tested. These performance claims come from the company and have not been peer reviewed.
In July 2026 Isomorphic and Google DeepMind published a joint statement on "bioresilience" that names the potential misuse of AI among the reasons for vigilance against biological threats, the Dual-use research of concern concern that also runs through debate over Engineered pandemics.
In October 2026 Isomorphic, Google DeepMind and Meta together committed $300 million to the Virtual Biology Initiative, an effort led by Biohub to generate open datasets for predictive models of cells. Isomorphic calls itself a founding member. The partners valued the whole commitment at $1.8 billion, but that total includes Biohub's $500 million from April, more than $500 million that the US Department of Energy plans to spend over five years, and datasets built with more than $500 million of earlier NIH funding. The release gives no breakdown of the $300 million among the three companies.7 Biohub's head of science told Reuters that data paid for by commercial funders would be embargoed, for an unstated period in which the funders can work on it, before public release.8
Alphabet funded the company alone until March 2025, when it raised $600 million in its first outside round, led by Thrive Capital with participation from GV and Alphabet.19 In May 2026 it announced a $2.1 billion Series B, again led by Thrive. Alphabet and GV took part, and the new investors were MGX, Temasek, CapitalG and the UK Sovereign AI Fund. The announcement gave no valuation.10 The company said the money would go to developing the engine, hiring and moving its pipeline toward the clinic. The external total is comparable to the roughly $3 billion reported as committed at launch to Altos Labs.
The company's clinical schedule has slipped. In 2025 Hassabis said Isomorphic would have AI-designed drugs in clinical trials by the end of that year. At Davos in January 2026 he said it expected its first trials by the end of 2026.9 As of October 2026 the company had not announced a first dosing. Its September 2026 essay said its preclinical data were helping it "gear up for clinical developments". At the Series B its executives described the approach as "fundamentally sound" and spoke of viable candidates, but none of that has been tested in a person.
The broader objection applies to the whole field. The largest cause of clinical failure is lack of efficacy, which better structures and affinity predictions do not directly address, and as of 2026 no AI-designed drug from any company has been approved. Jaderberg has himself said drug design will need "half a dozen more breakthroughs" of AlphaFold 3's magnitude.
What is and is not shownIsomorphic's published evidence is a peer-reviewed structure-prediction model and benchmark results reported by the company for a proprietary engine. No Isomorphic molecule has entered a clinical trial, the company has not publicly named a development candidate, and the partners' milestone payments depend on results not yet achieved.
The links to Google DeepMind raise a separate question. Both companies are run by the same chief executive, AlphaFold 3 was built jointly by the two, and the 2024 dispute over its release turned on how far commercial interests could limit access to a published result. The October 2026 data commitment poses a version of the same question: the resource is described as open, but the companies paying for part of it get a period of early access first.
The first test is whether an Isomorphic-designed molecule enters a clinical trial. The harder one is whether its candidates then succeed in phase 2 more often than the industry's historical rate, a comparison that needs many programmes and most of a decade. Calico Life Sciences, an earlier Alphabet life-sciences company, shows how long that kind of patience can run: more than ten years after its founding it had no approved drug.
Proponents argue that an engine which generalises across targets and drug types improves with every programme in a way a single asset cannot, and that proteins once considered undruggable become tractable. Critics answer that earlier computational tools were promoted in similar terms without moving approval rates in Cancer or elsewhere. Neither side can yet point to clinical data.
Whether such tools reach slower problems, such as interventions grounded in the Geroscience hypothesis, is a further question. Structure prediction learned from decades of experimentally solved structures; aging has no comparable body of measured outcomes and not even a validated surrogate endpoint. Cell-response data of the kind the Virtual Biology Initiative is meant to generate, and the simulation pursued under Human digital twins, are attempts to fill that gap, and neither yet exists at the scale the structural data did.
statementIsomorphic Labs. "Isomorphic Labs to appoint Max Jaderberg as President." Press release, 26 November 2025. ↩ ↩2
newsTaylor, P. "Isomorphic signs up Lilly, Novartis for $3bn AI drug hunt." pharmaphorum, 8 January 2024.↩The headline total is mostly contingent milestone payments; the upfront payments came to $82.5 million combined.
paperAbramson, J. et al. "Accurate structure prediction of biomolecular interactions with AlphaFold 3." Nature, 2024.↩Authors from Google DeepMind and Isomorphic Labs; as published, its code-availability statement said code was not provided and offered only a restricted non-commercial web server.
newsCallaway, E. "AI protein-prediction tool AlphaFold3 is now more open." Nature, November 2024.↩Code released for non-commercial use, with trained weights available only on request to academically affiliated scientists.
statementRoyal Swedish Academy of Sciences. "The Nobel Prize in Chemistry 2024." Press release, 9 October 2024.↩Names Hassabis and Jumper with Google DeepMind affiliations; the citation is for protein structure prediction, not for drug design.
preprintIsomorphic Labs Team. "Accurate Predictions of Novel Biomolecular Interactions with IsoDDE." Technical report, Zenodo, February 2026.↩A company technical report, not peer reviewed, describing proprietary models; the benchmark comparisons are the company's own.
statementBiohub. "International, cross-sector collaboration commits nearly $2 billion to build foundational data for AI models to predict and treat disease." Press release, 7 October 2026.↩The $300 million is given only as a joint figure for Google DeepMind, Isomorphic Labs and Meta, with no individual shares.
newsHu, K. "US government, Google join Zuckerberg-backed Biohub in $1.8 billion push for AI biology data." Reuters, 7 October 2026.↩Reports an embargo period for commercially funded data before public release; the length of the period is not stated.
newsReuters. "Google-backed Isomorphic Labs delays clinical trial timeline." 20 January 2026. ↩ ↩2
statementIsomorphic Labs. "Isomorphic Labs secures $2.1 Billion funding to scale its AI drug design engine." Press release, 12 May 2026.↩Names the lead and participating investors but discloses no valuation.