Isomorphic Labs
Isomorphic Labs is a London based AI drug design company founded in 2021 by Sir Demis Hassabis, who leads it as chief executive. It was launched from DeepMind as an Alphabet company, and with Google DeepMind it released AlphaFold 3 in May 2024. Its Isomorphic Labs Drug Design Engine (IsoDDE), introduced in February 2026, joins proprietary models for protein ligand structure prediction, binding site discovery, binding affinity and antibody antigen modeling with generative agents that design molecules. The company reports IsoDDE more than doubling AlphaFold 3's accuracy on the protein ligand systems least like its training data, reaching 2.3 times its rate of high fidelity antibody antigen predictions, and beating other deep learning methods on binding affinity. The engine is proprietary and runs only in the company's own and partnered programs.
Isomorphic Labs runs its own oncology and immunology programs alongside research collaborations with Novartis, Eli Lilly and Johnson & Johnson; none of its molecules has reached a clinical study yet. It raised $2.1 billion in Series B funding in May 2026, led by Thrive Capital with Alphabet and GV among the backers, after a $600 million round in 2025, and has offices in Cambridge, Massachusetts and Lausanne, Switzerland. With Google DeepMind it published a joint approach to bioresilience in July 2026 aimed at stopping misuse of its models.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
IsoDDE, the company's drug design engine, does the discovery work: proprietary models predict protein ligand structures, find binding sites from a protein's amino acid sequence alone, estimate binding affinity and model antibody antigen interfaces, while generative design agents search chemical space against a specification set by drug designers. Partnered programs with Novartis, Eli Lilly and Johnson & Johnson and the internal oncology and immunology pipeline all start from that engine. Isomorphic Labs sells no instrument, assay or software product alongside it.
Drug designers at Isomorphic Labs give IsoDDE a design specification: the biological target, the type of molecule and mode of modulation, the selectivity required, and properties such as solubility and cell permeability. A design agent then works autonomously for two to four days, generating molecules in batches and scoring them against that specification, after which the team selects several diverse compounds to synthesize and test in the lab. Partners run their own preclinical testing on the programs they take forward.
No rule is published for when a person overrides the agent or what confidence accompanies a prediction.
The drug design engine pairs proprietary foundation models for structure, binding, molecular properties and toxicity with generative design agents, and goes beyond AlphaFold 3, whose architecture Isomorphic Labs and Google DeepMind published in 2024. A February 2026 technical report gives IsoDDE results on public benchmarks (Runs N' Poses, FEP+ 4, OpenFE and the CASP16 binding affinity task) and on a 334 complex antibody antigen test set. The models draw on a curated collection of life science data the company calls its dataverse, along with health data from research studies.
The architecture, code and weights stay private, so the figures cannot be reproduced outside the company.
IsoDDE is built in house from proprietary models and runs across vast GPU fleets against the company's curated life science data, alongside health datasets obtained through agreements with third party organizations. The privacy notice names outside parties by role: Alphabet and Google affiliates that provide shared business operational services, service providers for IT, hosting and data security, and clinical research organizations.
No hosting provider, subprocessor list or rule on who can reach partner targets and chemistry is published.
Published results for IsoDDE come from the company's own runs on public benchmarks built by others, released in February 2026. On the Runs N' Poses generalization benchmark it more than doubles AlphaFold 3's protein ligand accuracy on the systems least like its training data; on a low homology set of 334 antibody antigen complexes it reaches 2.3 times AlphaFold 3's rate of high fidelity predictions (DockQ above 0.8) and 19.8 times that of Boltz-2; and on FEP+ 4, OpenFE and CASP16 it beats other deep learning methods on binding affinity and can exceed physics based FEP. The engine also predicted a recently discovered binding site on cereblon from the protein's sequence alone, and in a September 2026 example a design agent produced a molecule that modulated its target in laboratory testing. AlphaFold 3, the predecessor, appeared in peer reviewed form with Google DeepMind in 2024.
No molecule from the platform has yet entered a clinical study.
Protein structures and molecules are the engine's main inputs, and the company also uses health and genetic data in research: healthcare records, body scans, data from blood samples and questionnaire answers, drawn from biobanks, third party studies, clinical research organizations, laboratories and healthcare providers. Under its privacy notice this data trains the company's AI models, is processed in anonymized, de identified or pseudonymized form where possible, and may be kept indefinitely for research; the notice adds that some genetic data cannot be effectively anonymized.
Nothing says whether partners' confidential targets and chemistry also train the shared engine.
Isomorphic Labs Limited handles personal data under UK data protection law, including the GDPR and the Data Protection Act 2018. For health and genetic data used in research, its privacy notice sets out the categories (healthcare records, body scans, blood samples, questionnaires on medical history and mental health), where the data comes from, the special conditions relied on (scientific research in the interests of society, information a person made public, or explicit consent), safeguards, international transfers, retention and individuals' rights. Partners' confidential targets and chemistry fall under collaboration agreements rather than the notice.
The company makes no statement on HIPAA or business associate status.
Personal and research data at Isomorphic Labs sit behind multifactor authentication, data encryption, vulnerability and security management, access limited to staff with a strict business need and checks on service providers' security, and, where appropriate, Trusted Research Environments that restrict access, prevent downloads and monitor all activity around sensitive research data. What needs protecting includes partners' confidential targets as well as research health data.
No security certification, trust center or independent audit report is published.
IsoDDE proposes and evaluates molecules inside drug discovery programs and makes no diagnostic or treatment recommendation, so no device clearance applies. Regulation will attach to the drug candidates as they move toward the clinic: the company says its preclinical data is helping it gear up for clinical development, and Dr Ben Wolf serves as chief medical officer.
How model outputs would be documented if they supported a regulatory submission has not been stated.
Isomorphic Labs and Google DeepMind published a joint approach to bioresilience in July 2026: more than 15 partnerships with government bodies, biosecurity organizations and research groups to stop threat actors misusing their models, and a focused Isomorphic unit that deploys the drug design engine on medical countermeasures for governments and nonprofits during outbreaks. IsoDDE's headline structure results are reported separately for the protein ligand systems least similar to its training data, where AlphaFold 3 struggled to generalize.
No model card or analysis of how the engine performs across target classes is published.
Designs from IsoDDE go through synthesis and laboratory testing before the company advances them, and partners carry out their own preclinical studies before anything nears a patient. Any warranty, remediation or fee consequence for a failed design sits in private collaboration agreements with Novartis, Eli Lilly and Johnson & Johnson.
No accuracy figure on partner programs, and no remedy for a wrong prediction, is published.
Partners reach IsoDDE through research collaborations, and the engine works on protein sequences, structures and chemistry rather than patient records. Health data used in research, including healthcare records, arrives as datasets under agreements with third party organizations, research partners and healthcare providers.
No integration with partners' lab notebooks, laboratory information systems or research data platforms, and no data standard, is named.
The engine runs inside Isomorphic Labs, so partners bring targets to the company rather than installing anything. Isomorphic Labs Limited is registered at 280 Bishopsgate, London, with offices in Cambridge, Massachusetts and Lausanne, Switzerland, and its privacy notice allows personal data to be stored and processed outside the country where it was collected, including in the United States, relying on Google's Data Privacy Framework certification, UK adequacy decisions or standard contractual clauses.
Where partner data and models are hosted, and how one partner's work is kept apart from another's, is not stated.
Drug companies work with Isomorphic Labs through negotiated research collaborations, opened through a dedicated partnering contact. Novartis signed in January 2024 for small molecules against three challenging targets and expanded the deal in February 2025 with up to three more programs; Eli Lilly's collaboration covers small molecule programs; and Johnson & Johnson signed a collaboration across several targets and modalities in January 2026.
No price or standard deal terms are published.
IsoDDE covers protein ligand structure prediction, binding site discovery, binding affinity and antibody antigen modeling, and the company describes the engine as working across therapeutic areas and drug modalities, from small molecules to antibodies, peptides and molecular glues. The internal pipeline focuses on oncology and immunology, while partnered work spans small molecules with Novartis and Eli Lilly and small molecules and biologics with Johnson & Johnson.
Results by therapeutic area have not been reported.
Alternatives to Isomorphic Labs
Vendors in the same categories as Isomorphic Labs whose graded profiles look most alike, with a line on what each product does.
Similar is not the same, so check that each one does the job you need. No vendor pays for placement.
Biolojic Design uses machine learning to design antibodies that do more than block a target: epitope specific agonists and antagonists, conditional binders, and what it calls multibodies…
Earendil Labs uses AI to design antibody drugs and develops them through its own pipeline, licensing individual programs to large pharmaceutical companies.
Galux designs protein therapeutics with an AI platform that it pairs with the physics of how proteins fold.
LabGenius designs antibody drugs for cancer using a machine learning engine wired into its own robotic lab, and develops them through a mix of partnerships and a wholly owned pipeline.
Numerion Labs, formerly known as Atomwise, is a San Francisco AI drug discovery company developing small molecules for immune and inflammatory disease, led by chief executive Steve Worland.
Scala Biodesign makes software that redesigns proteins to work better, and sells it both as a cloud platform and as a design service.
Ordered by how closely the graded profiles match across the fifteen axes, inside the same categories. See how each axis is graded in the methodology.
Compared With
Each comparison carries a written verdict, the buyer conditions that favor each vendor, and a graded side by side. Pairs that cross a category boundary are grouped separately, and their verdicts state where the boundary sits rather than manufacturing a head to head.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Not applicable
|
Strategic research collaborations only. Reported structures use upfront payments plus success based milestones and royalties. | — | Not applicable. No software is delivered to the partner. | Vendor Published |
The engine is reached only through negotiated research collaborations with pharmaceutical companies. Reported deals with Eli Lilly and Novartis (2024, the latter expanded in 2025) and Johnson & Johnson (2026) carry potential milestone values specific to those partners. How rights to designed molecules, data and model outputs are split is not public.