AlphaFold's Drug-Design Spinoff Just Entered Human Trials

AlphaFold won its creators a Nobel Prize in 2024 for solving a problem biologists had chased for fifty years: predicting a protein's 3D shape from its amino acid sequence alone. The much harder commercial question was always whether that capability could actually design new drugs, not just describe existing proteins. In 2026, Isomorphic Labs — the Google DeepMind spinoff built specifically to answer that question — moved its first AI-designed drug candidates into human clinical trials.

What Isomorphic Labs actually does

Isomorphic Labs, led by DeepMind co-founder Demis Hassabis, uses AlphaFold's successors as the foundation for de novo drug design: rather than predicting the structure of a protein that already exists, the company's models predict how a novel drug molecule would need to be shaped to bind a specific disease-relevant target, then generate candidate molecules to fit. That's a different approach than editing an existing gene, the strategy behind CRISPR Therapeutics' cholesterol therapy — Isomorphic is designing an entirely new molecule rather than switching off a gene that's already there. The company has raised roughly $2.1 billion to build out this pipeline, and its stated ambition — solving "un-druggable" targets, proteins too structurally complex for conventional medicinal chemistry to address — is deliberately audacious.

What actually entered trials

The company's own announcements around its first clinical candidates have been notably light on specifics: no named disease target, no specific drug candidate identified publicly, and no concrete trial-phase details beyond describing a transition into "clinical testing." That's a meaningfully different posture than most biotech announcements, which typically lead with the target and mechanism. For a company whose entire premise is that its AI can do something conventional drug discovery can't, withholding the technical specifics of its first real-world test is a notable choice.

Withholding the technical specifics of its first real-world test is a notable choice for a company whose entire premise is that its AI does something conventional discovery can't.

Why "un-druggable" targets are the real test

The interesting claim isn't that AI can design drugs at all — AI-assisted drug discovery has been part of the pharmaceutical industry's toolkit for years. It's that Isomorphic's approach can address protein targets that have resisted conventional structure-based drug design entirely: proteins with shallow binding pockets, disordered regions, or shapes that don't offer an obvious place for a small molecule to attach. If AlphaFold-derived design tools can crack even a handful of those, it would be a genuinely new capability, not just a faster version of an existing one.

What "entering trials" doesn't mean yet

Entering Phase 1 testing means confirming a drug is safe in humans at various doses — it says essentially nothing about whether the drug actually works. The vast majority of drugs that enter Phase 1 trials, AI-designed or not, never reach approval; industry-wide success rates from Phase 1 to approval sit somewhere around 10%. Isomorphic Labs reaching this stage is a real, meaningful milestone for the company and the field, but it's the beginning of a yearslong evaluation process, not evidence the underlying approach works.

The honest read

This is a genuinely important test case for whether AI-native drug design produces better molecules than conventional methods, or just faster ones. Faster drug discovery that still fails at the same rate in clinical trials would be a real but modest win. Discovery that succeeds at a meaningfully higher rate, because AI can explore chemical and structural space human chemists can't, would be the actual breakthrough the field has been promising for a decade. 2026's trial data won't fully answer that question, but it's the first real data point.