Loading
What changed between revision 1 and revision 2, compared over the article's markdown source with its frontmatter — so a change to status, evidence or reversibility shows up as the edit it is.
Older · revision 1
Import· initial corpus import, not an edit
Initial import of content/isomorphic-labs.md — the filesystem corpus, unchanged. Not an edit.
15.6 kB
Revision 1
Revision 2
lines 7–13 → 7–136 unchanged lines not shown
categories: ["organizations", "foundations"]categories: ["organizations", "foundations"]tags: ["drug discovery", "machine learning", "alphafold", "structural biology", "google", "biotech"]tags: ["drug discovery", "machine learning", "alphafold", "structural biology", "google", "biotech"]summary: "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."summary: "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."updated: "2026-10-27"updated: "2026-10-29"------ ```infobox```infoboxlines 68–80 → 68–8054 unchanged lines not shown
### AlphaFold 3### AlphaFold 3 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.[^abramson2024]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.[^abramson2024] 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.[^callaway2024] 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.[^nobel2024]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.[^callaway2024] 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.[^nobel2024] ### The drug design engine### The drug design engine 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.[^isodde2026] 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 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.[^isodde2026] 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]] concern that also runs through debate over [[engineered-pandemics]].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]] concern that also runs through debate over [[engineered-pandemics]]. removed, struck through added, underlinedLine numbers count the serialised markdown of each revision, frontmatter included.
128 lines are the same in both revisions, and 111 of them are not shown. Neither revision here is necessarily the current article — read Isomorphic Labs for the text this wiki stands behind today.