Isomorphic Labs
AI drug design company spun out of Google DeepMind in 2021 and founded by Demis Hassabis, building on the AlphaFold structure prediction lineage. In February 2026 the company released IsoDDE, a unified drug design engine combining structure prediction, ligand binding, affinity prediction and antibody interaction modeling in a single pipeline. Reported performance is roughly double AlphaFold 3 accuracy on the hardest ligand binding cases where the target has under 20 percent sequence similarity to training data, and 2.3 times AlphaFold 3 on antibody antigen docking in the high fidelity regime. Unlike AlphaFold 1 through 3, IsoDDE is proprietary: no code, no weights, no public API and no peer reviewed publication, and Nature reported that the technical paper offers scant insight into how the results were achieved. Access is available only through pharma partnership. Named partners are Eli Lilly and Novartis, announced January 2024 with combined potential milestone value near 3 billion dollars, expanded with Novartis in February 2025, plus Johnson & Johnson. The company raised 600 million dollars in March 2025 led by Thrive Capital with GV and Alphabet participating. First in human trials for an AI designed candidate are targeted for end of 2026, a timeline that slipped from an earlier 2025 target.
Capability Axes
The model is the entire company. There is no instrument business, no assay platform and no legacy software franchise underneath it: what pharmaceutical partners are paying for is access to a drug design engine descended from AlphaFold. IsoDDE, released 10 February 2026, unifies structure prediction, ligand binding, affinity prediction and antibody antigen interaction modeling in a single pipeline, and the stated ambition is to reimagine drug discovery from first principles. On this axis the grade is unambiguous even where other axes are not.
Structurally the oversight is the partner's, and that is a real mechanism: designs generated by the engine enter Lilly's, Novartis's or Johnson & Johnson's own preclinical development processes, which are among the most rigorous validation environments in the industry, and nothing reaches a patient on the model's say so. Held at B because nothing vendor side is documented. No disclosure was located on how outputs are triaged internally, what confidence is attached to a prediction, or what the engine is known not to do well. The closed nature of the system compounds this: a partner can observe that a design failed but cannot inspect why the model proposed it.
The sharpest disclosure gap in this category, and it is notable precisely because of the lineage. AlphaFold 2 and 3 were published, and the public AlphaFold Database serves more than 3 million researchers across more than 190 countries with over 230 million predicted structures. IsoDDE reverses that posture entirely: no code, no weights, no public API and no peer reviewed publication, with Nature reporting that the technical paper offers scant insight into how the results were achieved and independent researchers left to guess at the method. Every performance figure, including roughly double AlphaFold 3 accuracy on the hardest ligand cases under 20 percent sequence similarity and 2.3 times on antibody antigen docking in the high fidelity regime, is vendor generated on vendor selected benchmarks and cannot currently be reproduced or contested by anyone outside the company. That is not a criticism of the science, which may well be as good as claimed. It is a statement that the claim is unfalsifiable from outside.
No molecule from this platform has entered a human being. First in human trials are targeted for the end of 2026, a date that already slipped from an earlier 2025 target, with candidates described as in IND enabling work across oncology and immunology. Partner programs are reported to have progressed from target identification to multiple preclinical candidates by early 2026. The evidence that does exist is commercial rather than clinical: roughly 3 billion dollars in potential milestone value across the Lilly and Novartis agreements, plus Johnson & Johnson. Applying the index precedent that scale does not substitute for evidence of benefit, deal value is a signal about what sophisticated partners believe, not a demonstration that the platform works. Buyers comparing platforms should hold this against peers with molecules in patients.
Not applicable in the provider sense and rated accordingly rather than penalized. The engine operates on protein structures, ligands and molecular interaction data, not on patient records. The confidentiality question that does arise is the partner's target and chemistry information, which is governed by the collaboration agreement.
Not applicable. The counterparties are pharmaceutical research organizations entering multi target discovery collaborations, not covered entities transferring protected health information.
No SOC 2, ISO 27001 or equivalent attestation was located and no trust center was found. Partners are entrusting their most sensitive undisclosed target information to a company held within Alphabet, and the absence of a published attestation means that diligence runs entirely through the collaboration agreement and whatever assurances the parent structure provides.
The lowest regulatory standing in this category, which is a statement of stage rather than of conduct. The design engine is not a regulated device and is not presented as one, correctly. At asset level no cleared IND, registered clinical trial or human dosing was located, with candidates described as in IND enabling work and first in human trials targeted for end of 2026. A buyer comparing platforms should be clear that the most technically celebrated engine in the category has, as of this review, no regulatory track record of its own to assess.
No AI governance framework or bias disclosure was located. The domain relevant question is whether the engine performs unevenly across target classes and chemical space, and the company's own headline metric hints at exactly this by reporting accuracy separately for cases below 20 percent sequence similarity to training data. Because the model is closed, no external party can characterize where it degrades, which converts an ordinary technical limitation into a governance one: the buyer cannot audit the boundary.
Not applicable. There is no provider workflow surface and no EHR touchpoint. Interoperability in the research sense is also constrained, since no public API or distributable software exists to integrate with.
The most restrictive access model in the category. There is no software to license, no API to call and no self service tier: the engine is available exclusively through negotiated pharmaceutical partnership, currently disclosed with Eli Lilly, Novartis and Johnson & Johnson. In practice a research organization cannot use this platform at all without a strategic collaboration of a scale only large pharmaceutical companies can execute. No hosting, tenancy or data residency terms were located, which follows from there being no deployment in the ordinary sense.
Deal shape is public and specific even though no rate card exists or could exist. Disclosed structures include the January 2024 Lilly and Novartis collaborations with combined potential milestone value near 3 billion dollars, reported as up to approximately 1.7 billion for Lilly and up to approximately 1.2 billion for Novartis, a February 2025 Novartis expansion adding up to three further research programs, a subsequent Johnson & Johnson agreement, and a 600 million dollar round in March 2025 led by Thrive Capital with GV and Alphabet participating. What is not disclosed is per program economics or what a partner actually receives for the money.
Broad on modality, concentrated on stage. IsoDDE spans small molecule work and antibody design in one pipeline, with reported strength on the CDR H3 loop that is the hardest region of an antibody to predict, which is genuinely wider modality coverage than most peers offer from a single engine. Disclosed therapeutic focus centres on oncology and immunology, with cardiovascular disease also reported. All of it sits at the design and preclinical stage, so coverage claims describe where the engine is pointed rather than where it has delivered.
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
|
Exclusive strategic partnership only. Reported structures use upfront payments plus success based milestones and royalties. | — | Not applicable. No software is delivered to the partner. | Vendor Published |
The access model matters more than the price here. There is no licence, no API and no self service tier, so a research organization cannot use this platform without negotiating a strategic collaboration at a scale only large pharmaceutical companies can execute. Disclosed agreements are with Eli Lilly, reported at up to approximately 1.7 billion dollars in milestones, Novartis at up to approximately 1.2 billion dollars and expanded in February 2025 with up to three additional research programmes, and Johnson & Johnson. Combined potential value across the Lilly and Novartis deals is reported near 3 billion dollars. What a partner receives for that money, and on what terms, is not public.