The gap between protein-structure prediction and an approved drug is where every AI pharma bet eventually has to prove itself. Isomorphic Labs is about to find out which side of that gap it's on.
By Shashi Bellamkonda · April 26, 2026
Isomorphic Labs built the best benchmark story in computational drug discovery. Clinical trials will test whether that story survives contact with human biology. For enterprise pharma IT teams, the platform bet now has a deadline.
For decades, discovering a new drug meant years of scientists manually testing thousands of compounds in a lab, hoping one of them would bind to the right protein in the body and trigger a useful effect. The odds were terrible and the costs enormous. Isomorphic Labs is betting that artificial intelligence can do much of that work on a computer before a single experiment runs in a lab.
The foundation is Google DeepMind's AlphaFold, an AI system that solved one of biology's hardest puzzles: predicting the three-dimensional shape a protein will fold into, based purely on its chemical sequence. That shape determines what a protein does in the body and, critically, where a drug molecule might attach to it to change its behavior. AlphaFold earned its creators the Nobel Prize in Chemistry in 2024. More than three million researchers across 190 countries now use it.
Knowing the shape of a protein, however, is not the same as knowing how to design a drug that fits it. Isomorphic Labs built its own system on top of AlphaFold to do that next step. The system, called IsoDDE, is designed to predict not just protein shapes but how tightly a candidate drug molecule will grip the target, what else in the body it might accidentally affect, and whether hidden attachment points exist on proteins that were previously thought to be unreachable.
Now Isomorphic Labs president Max Jaderberg has confirmed the company is preparing to bring its first AI-designed drug candidates into human trials, focused on cancer and immune system diseases. Speaking at the WIRED Health conference in London on April 16, Jaderberg said the molecules are designed to be highly effective at lower doses, which typically means fewer side effects. This is the moment the company has been building toward since it was spun out of Alphabet in 2021.
The timeline deserves scrutiny.
The slip is not a small detail
DeepMind chief executive Demis Hassabis said in 2025 that Isomorphic would have AI-designed drugs in human clinical trials by the end of that year. At the World Economic Forum in Davos in early 2026, he revised that to end of 2026. Now, in late April 2026, Jaderberg says the company is gearing up, without providing a specific date. That is three public statements across eighteen months, each one pushing the definitive test further out.
Drug development slips happen. Getting a new molecule ready for human testing involves regulatory filings, manufacturing preparation, safety reviews, and patient recruitment protocols, all of which take longer than anyone outside the industry expects. None of that reflects poorly on the underlying science. But technology leaders inside large pharmaceutical companies who are making budget and infrastructure decisions based on when AI-designed drugs will reach the clinic need to account for the pattern. A story built on "wait until we get into trials" accumulates a credibility cost each time the date moves.
"AlphaFold maps the terrain. IsoDDE is supposed to build the roads. Human trials will tell you whether the roads go anywhere."
This is a different kind of drug company
Isomorphic is not trying to become a traditional pharmaceutical company. Its model is to handle the computational work of drug design, and then hand candidates off to established pharma partners who run the clinical trials and bring the drug to market. It has formal partnerships with Eli Lilly, Novartis, and Johnson & Johnson structured along those lines.
If that model works, the implications for pharma technology spending are significant. Traditional drug discovery involves enormously expensive computer simulations that require specialized scientific expertise to run and interpret. IsoDDE is reported to deliver comparable results in far less time and at a fraction of the cost. For a chief information officer or chief technology officer at a large drug company, that is the number worth stress-testing, not the Nobel Prize backstory.
Benchmarks are not patients
The internal test results Isomorphic has published are strong. IsoDDE outperforms AlphaFold 3 on the hardest prediction challenges. It successfully identified a hidden attachment site on a protein called cereblon, predicting the location from the protein's chemical sequence alone, before any lab experiment had found it. That is a meaningful demonstration of capability.
The gap between a strong computer test and a drug that works in a human being is where most drug discovery efforts fail. A molecule can look perfect on a computer and behave completely differently inside a living body. The biology is messier than any model. Drug candidates fail in human trials at high rates across the entire industry, regardless of how they were discovered. The honest question is whether AI-designed compounds fail less often, and for different reasons, than traditionally discovered ones.
Isomorphic will not know the answer until it has human trial data. Neither will anyone it is trying to partner with or sell to.
If you are a technology leader at a large pharmaceutical company, the right question to ask Isomorphic Labs right now is not about AlphaFold or IsoDDE benchmarks. It is: what is your plan for the data that comes back from your first clinical trials, and how quickly will that data loop back into the model? A computational drug design engine that cannot learn fast from human trial outcomes will not change pharma economics. It will just be an expensive way to generate the same Phase 1 failures the industry already has. Press for that answer before you commit infrastructure to the platform story.
Mullin, Emily. "AI-Designed Drugs by a DeepMind Spinoff Are Headed to Human Trials." WIRED, 24 Apr. 2026, wired.com.
Isomorphic Labs. "The Isomorphic Labs Drug Design Engine Unlocks a New Frontier Beyond AlphaFold." Isomorphic Labs, 10 Feb. 2026, isomorphiclabs.com.
Isomorphic Labs. "A Unified Drug Design Engine for a New Era of Discovery." Isomorphic Labs, isomorphiclabs.com.
