The directions that can currently be confirmed are: a peer-reviewed structure model shared with Google DeepMind (AlphaFold 3), a closed drug-design engine whose performance is company-reported against an external benchmark (IsoDDE), three named pharmaceutical partnerships, and a $2.1 billion round raised to push a pipeline toward the clinic. Every capability claim past AlphaFold 3 rests on the company's own evaluation of a model nobody outside can run. Nothing in the public record names a drug candidate, a target, or a trial.
Scope note: this page tracks the public record only: the company's own releases and technical reports, the peer-reviewed literature, and third-party reporting. No inside information. It records how far public material goes, not how well the company is doing.
Overview
| Direction | What can be confirmed | Latest evidence |
|---|---|---|
| 1. Structure prediction | AlphaFold 3 is published in Nature and its database is widely used | Nature, 2024 |
| 2. Drug-design engine | IsoDDE claims to more than double AF3 accuracy on an external generalisation benchmark | Company report, 2026-02-10 |
| 3. Partnerships | Novartis, Eli Lilly and Johnson & Johnson; Novartis expanded after a year | Expanded 2025-02 |
| 4. Capital | $2.1B Series B led by Thrive, with Alphabet, GV and four new investors | 2026-05-12 |
| 5. Pipeline | A pipeline is referred to; no candidate, target or trial is named | No primary source |
1. Structure prediction: the part that was peer-reviewed
What can be confirmed: AlphaFold 3 was published in Nature in 2024, jointly with Google DeepMind. It is the only part of this record that has been through peer review, and the only part an outside group can examine on its own terms.
Isomorphic states that the AlphaFold Protein Structure Database has been used by "over 3 million researchers in more than 190 countries." That is a usage figure for the database, not evidence about the drug-design engine built after it.
Boundary: structure and interaction prediction. A structure is an input to drug discovery, not an outcome of it. Confidence: high (peer-reviewed publication).
2. Drug-design engine: IsoDDE
What can be confirmed: on February 10, 2026 the company published a technical report introducing the Isomorphic Labs Drug Design Engine, described as "a unified computational drug-design system, progressing beyond AlphaFold 3 in its predictive accuracy." Three claims are made:
- It "more than doubles the accuracy of AlphaFold 3" on the hardest split of the Runs N' Poses benchmark (Škrinjar et al. 2025), which was built by others specifically to test generalisation to novel pockets and ligands.
- It "predicts small molecule binding-affinities with accuracies that exceed gold-standard physics-based methods at a fraction of the time and cost."
- It can "accurately identify novel binding pockets on target proteins using only the amino acid sequence as input."
The detail worth holding onto is which benchmark. The company did not grade itself on a test of its own construction; Runs N' Poses is external, and it was designed to probe exactly the generalisation gap the report says AlphaFold 3 left open. That choice is meaningful, and it is also as far as the independence goes: the evaluation was run by the company, on a model nobody outside can run, and reported in a company document rather than a reviewed venue.
Boundary: prediction (structure, affinity, pocket identification). Nothing here is a molecule that worked in an animal or a person. Confidence: medium for the relative claim (an external benchmark, self-run, unreplicated); low for anything downstream of it.
Open leads: whether IsoDDE is submitted for peer review; whether any third party is given enough access to reproduce the Runs N' Poses result; how much of the gap over AlphaFold 3 survives on targets chosen by someone else.
3. Partnerships: three named pharmaceutical companies
What can be confirmed: the company names three collaborations on its partnerships page.
- Novartis: announced January 2024, initially three "particularly challenging" small-molecule targets. Expanded in February 2025 by up to three additional research programs, with Novartis's president of biomedical research quoted on "the exploration of new chemical spaces that would be unavailable to probe through traditional methods."
- Eli Lilly: small-molecule therapeutics against an undisclosed number of undisclosed targets.
- Johnson & Johnson: multi-target and cross-modality.
What an expansion is evidence of. Novartis renewed and widened the scope after roughly a year of work. That is a partner with access to the programme deciding to buy more of it: a meaningful signal about demand, and about what one informed customer saw. It is not a clinical result, and the quoted claim about new chemical space is a partner's characterisation, not a published comparison.
Boundary: commercial validation. Confidence: high that the partnerships and the expansion exist (company-stated, named partners, quoted executives); none of that transfers to efficacy.
Open leads: whether any partner discloses which compounds came out of IsoDDE and which came from its own chemists: the provenance question that would make the engine's contribution legible.
4. Capital: $2.1 billion Series B
What can be confirmed: announced May 12, 2026 from London. $2.1 billion, led by Thrive Capital, with existing backers Alphabet and GV and new investors MGX, Temasek, CapitalG and the UK Sovereign AI Fund. The stated uses are continued development and deployment of IsoDDE, "accelerating and expanding its pipeline of therapeutic programs towards the clinic," and hiring across AI, engineering, drug design and clinical roles.
The round is evidence about capital and conviction, which is a real thing to know about a company whose costs are about to become clinical. It is not evidence about molecules. Alphabet's president is quoted describing "the early promise of the technology platform." The word is early, in the company's own release.
Boundary: financing. Confidence: high (company press release, named investors, dated).
5. Pipeline: what is not named
What can be confirmed: the funding release refers to "its pipeline of therapeutic programs" and to moving that pipeline "towards the clinic." That is the whole of it. The release names no candidate, no target, no indication, no trial and no date, and the word trial does not appear in it.
Secondary coverage has reported an expectation of first human dosing before the end of 2026. That claim is not in any primary source found here, and this page does not carry it.
Boundary: there is no disclosed clinical object to have a boundary around. Confidence: low (a pipeline is referred to and nothing in it is identified).
Open leads: a named development candidate; an IND filing; a registration on a trial registry. Any one of those would move this company from a prediction story to a clinical one, and none has appeared.
What could not be found
- No peer-reviewed publication for IsoDDE. The February report is a company document.
- No independent evaluation. No third party has reported running IsoDDE or reproducing its benchmark result.
- No named drug candidate, IND or trial registration.
- No clinical timeline in any primary source. The Series B release, which would be the natural place for one, does not give a date and does not mention trials.
- No provenance disclosure inside partner programmes: nothing that separates what the engine produced from what a partner's medicinal chemists produced.
- No model, weights or evaluation code. The distinguishing assets are closed, which is the stated business model rather than an omission.
Method and scope
Built from the company's own pages and technical report, its May 2026 press release, the AlphaFold 3 paper, and search for coverage. Every figure quoted above was read at its source. Where a claim appeared only in secondary coverage (the $2.6 billion committed-capital figure and the end-of-2026 dosing expectation), it was left out rather than repeated, because the primary documents do not carry it.
Company-reported performance is labelled as such throughout. The distinction this page tries to hold is between what has been reviewed (AlphaFold 3), what has been measured by the company on someone else's benchmark (IsoDDE), what has been bought (the partnerships), and what has only been referred to (the pipeline).
Primary sources
Sources
- Isomorphic Labs technology · Company
- IsoDDE · Technical report
- AlphaFold 3, Nature · Paper
- Partnerships · Commercial