Isomorphic Labs
Last reviewed
Sources
26 citations
Review status
Source-backed
Revision
v6 ยท 4,422 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
26 citations
Review status
Source-backed
Revision
v6 ยท 4,422 words
Add missing citations, update stale details, or suggest a clearer explanation.
Isomorphic Labs is a London-based artificial intelligence company that designs small molecule and antibody drugs using deep learning models derived from the AlphaFold family of protein structure predictors. The company was incorporated on February 24, 2021 and announced publicly on November 4, 2021 as a spin-out of Google DeepMind, structured as a sister company under Alphabet rather than as a DeepMind subsidiary. It is founded and led by Demis Hassabis, who serves as chief executive while simultaneously running Google DeepMind and who shared the 2024 Nobel Prize in Chemistry with John Jumper and David Baker for the original AlphaFold work that underpins Isomorphic's technology stack. By May 2026 the company had raised roughly $2.7 billion in external equity (a $600 million Series A in 2025 and a $2.1 billion Series B in 2026) and signed drug-discovery partnerships with Eli Lilly, Novartis, and Johnson & Johnson.[1][2]
Isomorphic occupies a distinct position in the AI drug discovery sector. Where competitors such as Recursion Pharmaceuticals, Insilico Medicine, BenevolentAI, and Exscientia generally combine machine learning with high-throughput phenotypic screening, knowledge graphs of biomedical literature, or generative chemistry conditioned on protein sequences, Isomorphic builds its platform around explicit structural prediction of protein-ligand and protein-protein complexes. The thesis, articulated by Hassabis and former president Colin Murdoch in 2021, is that if the same deep learning paradigm that solved the protein folding problem can be extended to the full chemistry of drug interactions, the median ten-to-fifteen year drug discovery and pre-clinical development timeline can be compressed to a fraction of that figure. As of May 2026 the company has raised approximately $2.7 billion in external capital across two rounds, signed milestone-bearing partnerships with Eli Lilly and Novartis valued at up to $2.9 billion in aggregate, added a cross-modality research collaboration with Johnson & Johnson in January 2026, co-authored the AlphaFold 3 paper with Google DeepMind, released the IsoDDE drug design engine in February 2026, and is preparing to dose its first patient with an AI-designed compound by the end of 2026.[3][4][5]
The immediate trigger for Isomorphic Labs was the November 2020 release of AlphaFold 2 at the fourteenth Critical Assessment of Structure Prediction (CASP14) competition, which produced protein structure predictions of accuracy comparable to experimental crystallography and was widely described as the resolution of a fifty-year grand challenge in computational biology. DeepMind followed CASP14 by releasing predicted structures for the entire human proteome and, in partnership with the European Bioinformatics Institute, for more than two hundred million proteins through the AlphaFold Protein Structure Database. Internally, Hassabis and Colin Murdoch began to consider how the same modelling approach might be applied to small molecule and biologic design.[6][7]
Drug discovery is one of the most economically significant translation problems in biology. Industry estimates put the average capitalised cost of bringing a new molecular entity to market at $2 to $2.6 billion and the median timeline at twelve to fifteen years from target identification to regulatory approval, with roughly ninety percent of candidates failing in trials. Hassabis and Murdoch argued that better physical models of how candidate drugs bind to and modulate protein targets could substantially raise the prior probability of clinical success.[8]
Isomorphic Labs was incorporated in England and Wales as Isomorphic Laboratories Limited on February 24, 2021 and announced publicly on November 4, 2021. The structure chosen was an Alphabet subsidiary parallel to DeepMind rather than a DeepMind subsidiary, allowing the company to take outside capital, sign commercial contracts with pharmaceutical partners, and operate under healthcare-specific governance. Hassabis was named founder and CEO of both DeepMind and Isomorphic, an unusual dual-CEO arrangement.[9]
Isomorphic Labs is wholly owned by Alphabet through a chain of intermediate holding companies and operates from a headquarters in the Kings Cross district of London, sharing a building with Google DeepMind. The company opened a second European research site in Lausanne, Switzerland in December 2022, taking advantage of the local cluster of computational chemistry and machine learning talent associated with the Ecole Polytechnique Federale de Lausanne, and added a North American office in Cambridge, Massachusetts in 2024 to be closer to the Boston biotech ecosystem and to several existing pharmaceutical partners.[10]
Leadership has evolved several times as the company has scaled. The initial executive team announced in 2021 included Murdoch as president, Miles Congreve as chief scientific officer, and a small cadre of senior hires drawn from both DeepMind and the pharmaceutical industry. In May 2024 the company appointed Max Jaderberg, a longtime DeepMind research scientist who had been a senior contributor to the AlphaFold programme, as chief AI officer with responsibility for the core modelling roadmap. Pamela Carroll, recruited from a senior operating role at a large biotech, was named chief operating officer in 2024, and Sergei Yakneen joined as chief technology officer with responsibility for the production engineering and platform infrastructure.
In December 2025 the company announced that Jaderberg would succeed Murdoch as president effective January 1, 2026, with Murdoch transitioning back to a strategic role at Alphabet. The change reflected the shift from research stage to clinical and commercial stage. Hassabis remains chief executive. John Jumper, who led the AlphaFold 2 team at DeepMind and is corresponding author on the AlphaFold 3 paper, is not formally part of Isomorphic's leadership but is a frequent collaborator and was named co-recipient of the 2024 Nobel Prize in Chemistry alongside Hassabis.[11]
| Role | Name | Background |
|---|---|---|
| Founder and CEO | Demis Hassabis | Co-founder and CEO of Google DeepMind; 2024 Nobel laureate in Chemistry |
| President | Max Jaderberg | Former DeepMind research scientist and Isomorphic chief AI officer |
| Chief scientific officer | Miles Congreve | Former senior medicinal chemistry leader, Sosei Heptares |
| Chief operating officer | Pamela Carroll | Biotech and pharma operations executive |
| Chief technology officer | Sergei Yakneen | Engineering leader, formerly at Roche and Microsoft |
| Strategic advisor | Colin Murdoch | First president of Isomorphic Labs and chief business officer at Google DeepMind |
For its first three and a half years Isomorphic operated entirely on capital provided by Alphabet, an arrangement that allowed the company to grow to several hundred employees, sign two large pharmaceutical partnerships, and co-develop AlphaFold 3 without raising external equity. The first external round closed on March 31, 2025 and was announced publicly on April 1, 2025. Thrive Capital, the New York firm founded by Joshua Kushner that is also a major investor in OpenAI, led a $600 million round with participation from GV (the venture arm of Alphabet) and a follow-on investment from Alphabet itself. The round was structured as a Series A despite the company's relatively mature stage because no prior priced round had been completed. Wilson Sonsini Goodrich and Rosati served as legal counsel.[12]
The Series A was widely interpreted as an external validation of the company's technical platform and a precursor to a larger round once a clinical asset materialised. In May 2026, less than fourteen months after the Series A, Isomorphic announced a Series B of $2.1 billion, again led by Thrive Capital, with participation from Temasek of Singapore, MGX of the United Arab Emirates, CapitalG (Alphabet's growth investment arm), the United Kingdom's Sovereign AI Fund, and follow-on investment from Alphabet and GV. The $2.1 billion Series B was the largest disclosed financing round ever raised by an AI drug discovery company and one of the largest private rounds in the European biotech sector, bringing total external capital to roughly $2.7 billion. The company stated that proceeds would be used to power its AI drug design engine, scale the business globally, and progress its drug candidate pipeline toward the clinic, including hiring across AI research, engineering, drug design, and clinical functions.[13][14]
The Series B announcement included statements from Alphabet leadership and from Isomorphic's own executives. Ruth Porat, president and chief investment officer of Alphabet, said the company's trajectory was encouraging and that the funding would be used to bring interventions to market with greater speed. Hassabis framed the raise as a step toward the company's stated mission of solving all disease, saying the capital would let Isomorphic build out its drug design engine at scale. Jaderberg, by then president, said the drug design engine "works" and was giving the company a repeatable way to design new medicines for a wide range of diseases.[24]
| Round | Date announced | Amount | Lead investor | Other participants |
|---|---|---|---|---|
| Internal funding | 2021 to 2025 | Undisclosed | Alphabet | None disclosed |
| Series A | April 1, 2025 | $600 million | Thrive Capital | GV, Alphabet |
| Series B | May 12, 2026 | $2.1 billion | Thrive Capital | Temasek, MGX, CapitalG, UK Sovereign AI Fund, Alphabet, GV |
The Series B implied valuation was not publicly disclosed but multiple press reports placed the post-money figure in the range of $9 to $11 billion, making Isomorphic Labs one of the most valuable privately held AI companies in Europe.
Isomorphic Labs operates a hybrid business model that combines internal therapeutic programs (where the company owns the asset and bears the development risk) with collaboration agreements where it provides AI-driven target selection, hit identification, and lead optimisation services to a pharmaceutical partner in exchange for upfront payments, milestone payments, and royalties on any resulting commercial product. As of May 2026 the company has disclosed four significant partnerships.
The first partnership, with Genentech (a subsidiary of Roche), was disclosed in October 2022 and covered multiple targets across an undisclosed therapeutic area. Financial terms of the Genentech collaboration have never been made public. The agreement gave Isomorphic an early opportunity to demonstrate its platform on real pharma-grade campaigns and provided the company with operational learning about how to structure its services for a partner workflow.
The second and third partnerships were announced together on January 7 and January 8, 2024 in advance of the JPMorgan Healthcare Conference. The Eli Lilly agreement involved a $45 million upfront payment, eligibility for up to $1.7 billion in performance-based milestone payments, and tiered royalties up to low double digits on net sales of any commercial product. The collaboration focused on small molecule therapeutics against multiple undisclosed targets. The Novartis agreement was structured similarly, with a $37.5 million upfront payment, up to $1.2 billion in milestone payments, and tiered royalties from mid-single up to low double digits on net sales, covering three undisclosed targets. The combined headline value of roughly $2.9 billion in milestones and upfront payments across the two deals (about $3 billion as reported by industry press) made Isomorphic the most commercially active of the AI-native drug discovery companies as of early 2024.[15][16]
The fourth partnership, announced on January 20, 2026, is a cross-modality, multi-target research collaboration with Johnson & Johnson, conducted through its Janssen Biotech entity. Unlike the small-molecule-focused Lilly and Novartis deals, the Johnson & Johnson collaboration spans several modalities including small molecules, antibodies, peptides, and molecular glues, and is aimed at targets that have historically been difficult to drug. Under the agreement Isomorphic handles in silico compound design and prediction using its AI platform, while Janssen Biotech leads experimental validation and downstream development. Financial terms and the specific therapeutic categories were not disclosed. Isomorphic chief research officer Sarah Skerratt described the work as aimed at unlocking previously inaccessible areas of biology rather than simply speeding up existing pipelines.[25][26]
| Partner | Announced | Upfront | Milestone potential | Scope |
|---|---|---|---|---|
| Genentech (Roche) | October 2022 | Undisclosed | Undisclosed | Multiple undisclosed targets |
| Eli Lilly | January 7, 2024 | $45 million | Up to $1.7 billion | Small molecule therapeutics, multiple targets |
| Novartis | January 8, 2024 | $37.5 million | Up to $1.2 billion | Small molecule therapeutics, three targets |
| Johnson & Johnson (Janssen Biotech) | January 20, 2026 | Undisclosed | Undisclosed | Cross-modality (small molecules, antibodies, peptides, molecular glues), multiple targets |
The partnerships are non-exclusive with respect to therapeutic areas and do not constrain Isomorphic from pursuing internal programs in the same indication classes. Both Eli Lilly and Novartis received the right to nominate additional targets within the collaboration scope, and both agreements include provisions for joint research committees that review program progress on a quarterly basis.
Isomorphic Labs' modelling effort is descended directly from the AlphaFold programme at Google DeepMind but has diverged substantially in scope and architecture. AlphaFold 2, released in 2020 and 2021, was a static protein structure predictor: given an amino acid sequence, it produced a single most likely three-dimensional conformation of the folded protein. AlphaFold 2 used a custom Evoformer block operating on multiple sequence alignments together with a structure module that produced atomic coordinates. The model was trained on the Protein Data Bank, an open archive of approximately two hundred thousand experimentally determined protein structures.
AlphaFold 3, co-authored by Isomorphic Labs and Google DeepMind and published in Nature on May 8, 2024 (volume 630, pages 493 to 500), generalised the modelling target beyond single protein chains to include the joint structure of complexes containing proteins, nucleic acids, small molecule ligands, ions, and post-translationally modified residues. The architecture replaced the Evoformer with a simpler Pairformer and introduced a diffusion-based generative head that directly generates atomic coordinates. Reported accuracy improvements were dramatic: protein-ligand interaction accuracy exceeded that of state-of-the-art docking tools, antibody-antigen prediction accuracy was substantially higher than AlphaFold-Multimer v2, and protein-nucleic acid interaction accuracy was higher than specialist predictors.[17]
The AlphaFold 3 release was accompanied by a free public AlphaFold Server hosted by Google DeepMind that allows non-commercial researchers to submit jobs containing proteins, nucleic acids, ligands, ions, and modified residues. The server uses a slightly restricted version of the AlphaFold 3 model and applies daily query limits per user. Source code for AlphaFold 3 was not initially released with the paper, drawing criticism from segments of the scientific community and an editorial response from Nature about its code availability policies. DeepMind ultimately released the AlphaFold 3 inference code on November 11, 2024 for non-commercial use, with model weights provided on application.[18]
On February 10, 2026 Isomorphic Labs released a twenty-seven-page technical report describing the Isomorphic Labs Drug Design Engine, abbreviated IsoDDE. IsoDDE is a unified neural architecture trained to perform four principal tasks in a single model: protein-ligand structure prediction, antibody-antigen interface modelling, binding affinity estimation, and ligandable pocket identification. The report positioned IsoDDE as a substantial advance beyond AlphaFold 3, with the model designed specifically for drug discovery rather than the more general biomolecular interaction problem that AlphaFold 3 addresses.[19]
On a held-out generalisation benchmark for protein-ligand structure prediction the report stated that IsoDDE more than doubles the accuracy of AlphaFold 3. On a held-out test set of 334 novel antibody-antigen complexes IsoDDE achieved 39 percent accuracy in the high-fidelity regime defined as DockQ greater than 0.8, compared with 17 percent for AlphaFold 3 and 2 percent for Boltz-2, a 2.3-fold and 19.8-fold improvement respectively. On binding affinity estimation IsoDDE exceeded the accuracy of gold-standard physics-based free energy perturbation methods at a small fraction of the computational cost. On pocket identification IsoDDE was reported to recapitulate the recent experimental discovery by Dippon et al. of a novel cryptic binding site on the E3 ligase cereblon, identifying both the canonical and the cryptic pockets using only the amino acid sequence as input.[20]
The IsoDDE release drew commentary from senior figures in computational biology. Mohammed AlQuraishi, an associate professor of systems biology at Columbia University and a long-standing observer of the AlphaFold programme, described IsoDDE as a major advance on the scale of an AlphaFold 4. Unlike the AlphaFold 3 release, IsoDDE was not accompanied by a code release or a server; the model is internal infrastructure for Isomorphic and for its partners and is not available to the broader research community.[21]
| Date | Milestone | Significance |
|---|---|---|
| November 2020 | AlphaFold 2 wins CASP14 | Provides scientific basis for spin-out thesis |
| November 4, 2021 | Isomorphic Labs announced | Public launch as Alphabet subsidiary |
| July 2022 | AlphaFold Protein Structure Database expanded to 200M proteins | Established protein structure as solved problem |
| May 8, 2024 | AlphaFold 3 published in Nature | First joint Isomorphic and DeepMind output, extends model to small molecules |
| May 2024 | AlphaFold Server released | Free public access to AlphaFold 3 for non-commercial research |
| November 11, 2024 | AlphaFold 3 code released under non-commercial licence | Resolved code availability controversy |
| February 10, 2026 | IsoDDE technical report released | Dedicated drug design model surpasses AlphaFold 3 |
Isomorphic Labs has consistently declined to provide detailed disclosure of its internal pipeline, citing competitive sensitivity. The company has stated publicly that it is running programs in oncology and immunology among other therapeutic areas, and has indicated that it is pursuing both small molecules and antibody-based modalities. Detailed target lists, mechanisms of action, and indication-level disclosures have not been provided.
Clinical progress has been a focal point of public attention. Hassabis stated in interviews in early and mid-2024 that the company expected to have an AI-designed drug entering human trials before the end of 2025. That timeline was not met. At the World Economic Forum annual meeting at Davos in January 2026 Hassabis revised the target, stating that the first patient dosing was now expected by the end of 2026. He clarified at Davos and in subsequent interviews that his earlier statements had referenced pre-clinical work, which encompasses laboratory studies and animal models, rather than clinical trials proper, which require regulatory approval and dosing of human subjects. As of the May 2026 Series B announcement no Isomorphic-designed compound had been administered to a human patient, and the company described its pipeline as approaching but not yet at investigational new drug application stage.[22]
| Stage | Status | Notes |
|---|---|---|
| Target identification | Active | Multiple oncology and immunology programs |
| Hit discovery | Active | Driven by IsoDDE structural predictions |
| Lead optimisation | Active | Reported lead candidates in oncology |
| Pre-clinical | Active | Animal model and toxicology work underway |
| Investigational new drug filing | Not yet filed | First IND filings expected during 2026 |
| Phase 1 clinical trial | Not yet initiated | First patient dosing targeted by end of 2026 |
| Phase 2 or later | Not yet initiated | No human efficacy data |
The delay from the original 2025 timeline drew muted but not dismissive commentary from industry analysts. The complexity of progressing a novel chemical entity through investigational new drug (IND)-enabling toxicology studies is widely understood to be irreducible regardless of how quickly a candidate is identified, and several industry observers noted that even a 2026 first-in-human dosing event would represent an unusually compressed timeline from company founding (2021) to clinical trial relative to historical industry norms of seven to ten years for de novo target campaigns.
Isomorphic Labs operates in a sector with several established competitors. Recursion Pharmaceuticals, founded in 2013 and headquartered in Salt Lake City, has built an industrial-scale phenotypic screening platform that perturbs cells with hundreds of thousands of genetic and chemical interventions and applies deep learning to derive disease hypotheses across many indications in parallel. Recursion went public in 2021 and merged with Exscientia in 2024. Insilico Medicine, founded in 2014, pioneered generative chemistry models for de novo small molecule design and brought a generatively designed DDR1 inhibitor to first-in-human dosing in 2022 using its Chemistry42 and PandaOmics platforms. BenevolentAI applies a knowledge-graph approach integrating published biomedical literature with experimental data, most notably identifying baricitinib as a candidate treatment for COVID-19 in early 2020.
The distinctive feature of Isomorphic Labs relative to these companies is the explicit centrality of structural prediction of binding geometry in its design loop. Where Recursion derives candidate hits from cellular phenotypes, Insilico from generative chemistry conditioned on protein sequences, and BenevolentAI from textual and graph-based associations, Isomorphic centres its workflow on predicted atomistic structures of protein-ligand complexes that are validated against experimental crystallography and cryo-electron microscopy data. The approach is closer in spirit to traditional structure-based drug design but operates at substantially larger scale and with substantially higher predictive accuracy than the docking and free-energy tools that have historically dominated that subfield. The trade-off is that Isomorphic's approach is most natural for targets that admit a single binding pocket and a well-defined functional readout, while phenotypic platforms can in principle uncover new biology even when target identity is unknown.
Academic and industry reception of Isomorphic Labs' technical output has been broadly enthusiastic. The 2024 Nobel Prize in Chemistry awarded jointly to Hassabis, Jumper, and David Baker is the highest-profile validation of the AlphaFold programme and, by extension, of the scientific premise underlying Isomorphic. The AlphaFold 3 paper has accumulated thousands of citations within its first two years, and the IsoDDE report drew positive technical commentary from academic computational biologists.
Criticism has tended to focus on three issues. The first is code availability: the original AlphaFold 3 paper was published without source code, which generated an open letter from a section of the structural biology community and an editorial from Nature acknowledging the issue. The eventual November 2024 code release under a non-commercial licence partially addressed the criticism but did not satisfy advocates of fully open scientific release. The second is the gap between technical capability and clinical evidence. As of May 2026 Isomorphic has demonstrated technical excellence but has not produced human efficacy or safety data for any internally designed compound, and several observers have noted that the historical correlation between pre-clinical predictive accuracy and clinical success is weaker than commonly assumed because of the role of unanticipated toxicity, pharmacokinetics, and disease biology that even excellent structural models do not address. The third is the question of independence from Alphabet: Isomorphic's continued reliance on Alphabet for core compute, sister-company collaboration with DeepMind, and partial financing through the GV and CapitalG arms of Alphabet has raised questions about the company's effective autonomy and the long-term direction it would pursue if Alphabet's strategic priorities shifted.
As of May 2026 Isomorphic Labs is one of the most heavily capitalised and technically credentialed AI-native drug discovery companies in the world. The Eli Lilly, Novartis, and Johnson & Johnson collaborations, the $2.7 billion of disclosed external equity capital, the Nobel-laureate-led management team, and the demonstrated technical advances embodied in AlphaFold 3 and IsoDDE place the company in a small group of clear leaders in the application of frontier AI to pharmaceutical research. The principal open question is whether the technical advances translate to clinical and ultimately commercial success on a timeline that justifies the capital and attention deployed. The expected first patient dosing event before the end of 2026 is the most important near-term milestone. If Isomorphic-designed compounds prove safe and efficacious in early-phase trials over the subsequent two to three years, the company will have established a new template for drug discovery. If they do not, the company will have demonstrated the limits of pure structural prediction as a route to therapeutic value and will face a more difficult conversation about its long-term business model.[23]