Isomorphic Labs is getting ready for the part of AI drug discovery that matters most: testing molecules in people. The Google DeepMind spinoff says it is preparing to move AI-designed drugs into human clinical trials, turning years of protein-structure progress into a live test of whether the technology can produce safer, more useful medicines.
The update came from Isomorphic Labs president Max Jaderberg at WIRED Health in London. WIRED reported that Jaderberg said the company is "gearing up to go into the clinic", though he did not give a specific start date. That is a more cautious timeline than earlier comments from CEO Demis Hassabis, who had previously pointed to trials by the end of 2025.
What happened
Isomorphic was spun out of DeepMind in 2021 to apply AI-first methods to drug discovery. Its work builds on AlphaFold, the protein-structure system that helped reshape biology and later earned Demis Hassabis and John Jumper the 2024 Nobel Prize in Chemistry.
The company now describes its core platform as the Isomorphic Labs Drug Design Engine, or IsoDDE. In a February technical update, Isomorphic Labs said IsoDDE more than doubles AlphaFold 3's accuracy on a challenging protein-ligand structure prediction benchmark, predicts small-molecule binding affinities faster than physics-based methods, and can identify novel binding pockets from amino acid sequence alone.
The pitch is simple: AlphaFold helped researchers understand molecular structures; IsoDDE is meant to help design drug candidates that can act on those structures. That moves the problem from prediction toward intervention.
Why it matters
Drug discovery is slow because promising ideas usually die in the lab, in animal testing, or in clinical trials. AI can make the search faster, but only if it produces molecules that behave well outside a model. Human trials are the first serious filter for that claim.
Isomorphic has a stronger starting point than most AI-drug-discovery startups. It has DeepMind's AlphaFold heritage, a $600 million funding round, and partnerships with Eli Lilly and Novartis. Its internal pipeline is focused on oncology and immunology, areas where better target selection and potency can matter a lot.
The company is also trying to generalise across targets and drug types rather than build one narrow model for one narrow disease. Its technology page frames IsoDDE as an engine that can work across multiple therapeutic areas and modalities, with much of the early design work happening in silico before lab validation.
What this means for AI in medicine
The near-term question is not whether AI can replace drug developers. It cannot. The question is whether AI can improve the hit rate before expensive clinical work begins.
That distinction matters. A molecule designed by AI still has to pass toxicology, dosing, manufacturing, regulatory review, and real patient outcomes. Even a successful first trial would not mean AI has solved drug discovery. It would mean the field has finally produced clinical evidence instead of only better benchmarks.
AlphaProteo shows why the excitement is not baseless. Google DeepMind has reported strong lab performance for AI-designed protein binders across seven target proteins, including a first successful AI-designed binder for VEGF-A and an 88% binding rate for candidates against the viral protein BHRF1. That is not the same as making a drug, but it is real wet-lab progress.
Our take
Isomorphic Labs is entering the hard phase. The company's AI systems look impressive on structure prediction, binding, and design benchmarks, but medicine is not a benchmark problem. Biology is noisy, patients vary, and safety can ruin otherwise elegant molecules.
Still, this is the right kind of milestone. AI drug discovery has had plenty of platform claims and partnership announcements. Human trials will force a cleaner answer: did the model help create a candidate worth testing, and did that candidate behave better than the old process would have predicted?
If Isomorphic can show early clinical signal, it will strengthen the case that AlphaFold-style systems are becoming practical drug design tools. If the first candidates disappoint, the lesson will be less dramatic but just as useful: AI can speed parts of discovery without removing the clinical risk that defines the industry.
