Max Jaderberg, president of Isomorphic Labs, says the company sells no model or API access and uses its drug design engine only to make drugs: its own programmes are in oncology and in immunology and inflammation, partners pay through upfront fees, milestones and royalties, and every programme is still deep in preclinical work, none yet in people.
Drugs, not model access
Jaderberg describes two lines of work. One is internal programmes: Isomorphic has the idea, pays for it, and takes it from basic science to molecules, patents and clinical trials, deciding only once the evidence is strong whether to license, partner or sell the drug itself. The other is partnerships with drugmakers such as Lilly, Novartis and Johnson & Johnson, which bring years of biology and drug ideas while Isomorphic is paid in upfront fees, milestones and later royalties. The record already held those deals and May's $2.1B Series B, raised to move programmes toward the clinic, but not the company saying it sells no model access.
Its own pipeline: oncology, immunology and inflammation, still preclinical
He says the engine works for any disease and for pills, antibodies and peptides alike, but carrying a clinical programme for years takes specialists in that disease, so the internal programmes stay in oncology and immunology and inflammation and other areas go to partners. Asked how far the company is from commercialisation, he says it is still early: it is designing molecules in preclinical work and proving them in the lab, in cells and in animal models before going to patients. That matches the record, which holds Isomorphic's deals, its rounds and the IsoDDE technical report it scored itself, and no clinical result.
Reasoning models propose, simulation models test first
Jaderberg argues that reasoning models can pull data together and propose many hypotheses, and that the hard part is choosing the few worth testing, since time to make molecules and grow cells is limited. So Isomorphic also builds what he calls simulation models or world models: models like AlphaFold that approximate the physical world, so experiments run first on the computer and the ones that reach the lab and the clinic are more likely to work. He puts the speed of recent maths results down to proofs being checkable on a computer, while an idea about a cell has to be tested on a real cell; his judgment is that breakthroughs in biology will come, not at once, and only together with real experiments.
What the interview does not show
The interview names no programme, target or candidate, gives no date for entering the clinic and shows no data. The roughly 90% failure rate in clinical trials is an industry figure he cites, and "all medicines in the future will be designed by AI" is his own judgment. He says Demis Hassabis now has more time for Isomorphic, which fits the record of Hassabis stepping down as DeepMind's CEO in August, but it is his account.