How a Google DeepMind Spin-off Hunts Hidden Drug Targets

Overview

Isomorphic Labs, the Google DeepMind spinout behind AlphaFold, has built an AI system called the Isomorphic Drug Design Engine (IsoDDE) that hunts for hidden binding sites on proteins where new medicines might attach. The company raised $2.1 billion in funding, one of the largest biotech rounds ever, and signed drug-discovery deals with Novartis and Eli Lilly. Its engine predicts not only where a drug molecule binds to a protein but how tightly it binds, including pockets that stay invisible until the right molecule arrives. The work shows AI moving past structure prediction toward the harder task of designing real drugs.

Key Takeaways

Isomorphic Labs spun out of Google DeepMind to turn AlphaFold's protein-folding breakthroughs into actual medicines, a step that needs far more than predicting a protein's shape.
Its Isomorphic Drug Design Engine (IsoDDE) handles three jobs: predicting a protein's structure, finding the pockets where drugs bind, and estimating how strongly a molecule sticks to its target.
IsoDDE found a 'cryptic pocket' on a protein called cereblon, a hidden site that opens only when the right molecule binds and stays invisible in the protein's resting state.
The system is designed to generalize to brand-new pockets far from its training data, where earlier AlphaFold models lose accuracy, opening the door to novel drug mechanisms.
Isomorphic raised $2.1 billion in funding and holds partnerships with Novartis and Eli Lilly worth billions in potential milestone payments.
A group leader at the company cautions that accurate structure modeling does not mean drug discovery is solved, since drug development stays slow and difficult.

Stats & Key Facts

#$2.1 billion raised by Isomorphic Labs, one of the largest biotech funding rounds on record.
#Up to $1.7 billion in potential milestone payments in the Eli Lilly partnership, plus $45 million paid upfront.
#Up to $1.2 billion in potential milestone payments in the Novartis partnership, plus $37.5 million paid upfront.
#3 main prediction tasks handled by IsoDDE: structure prediction, pocket identification, and binding affinity.
#2024 Nobel Prize in Chemistry awarded for the AlphaFold protein-folding work the company builds on.
#3 office locations supporting the buildout: London, Cambridge in Massachusetts, and Lausanne.

How a Google DeepMind Spin-off Hunts Hidden Drug Targets

Why AlphaFold's Nobel-Winning Work Was Not Enough to Design Drugs

Predicting a protein's shape solved one problem but left the harder drug-design challenge open.

AlphaFold2 was recognized with the 2024 Nobel Prize in Chemistry because it largely solved protein folding, the long-standing puzzle of how a chain of amino acids twists into a 3D shape. But proteins do not sit alone in the body. They constantly interact with other molecules, including nucleic acids, small drug-like molecules, ions, and other proteins.

AlphaFold3 added the ability to model those interactions inside a single framework, so one model could represent many cellular components at once. Even so, designing a drug asks more than knowing a shape. Researchers need to predict where a molecule binds, how it binds, how tightly it holds, and how it behaves alongside every other protein in the body.

Inside the Isomorphic Drug Design Engine and Its Three Prediction Jobs

IsoDDE is a single computational system built to answer several questions a drug designer needs.

›Structure prediction: where and how a candidate molecule, called a ligand, attaches to its target protein.
›Pocket identification: locating the cavities on a protein surface where a drug might fit.
›Binding affinity prediction: estimating how strongly a molecule grips its target, a key signal of whether it might work as a drug.
›The engine is meant to support many drug types, including small molecules, antibodies, molecular glues, and peptides.

How IsoDDE Found a Hidden Cryptic Pocket on the Cereblon Protein

The company's headline example involves a protein central to how cells dispose of unwanted proteins.

Cereblon is a well-studied protein in the targeted protein degradation pathway, the cellular machinery that marks damaged or disease-causing proteins for destruction. Some drugs already work by recruiting cereblon to tag harmful proteins for removal.

The interesting part is a 'cryptic pocket.' In the protein's resting, unbound state this cavity looks like it does not exist. It opens only when the right molecule arrives, which makes it nearly impossible to spot with standard structural analysis. IsoDDE correctly placed both the known binding site and the newly found cryptic site in their exact positions, matching results later confirmed and published in the journal Nature in January 2026.

Generalizing to Novel Pockets Where Earlier Models Lose Accuracy

The real test for a drug-design AI is performance on targets it has never seen.

Independent groups studying AlphaFold3 found a clear pattern: the further a protein pocket sits from the examples in the training data, the more the model's accuracy drops. That is a problem for drug discovery, because the most valuable new medicines often aim at never-before-targeted pockets and novel mechanisms of action.

Isomorphic says its engine is built specifically to hold up in those distant, unfamiliar regions. If an AI only performs well on pockets similar to what it has already seen, it offers little help in finding genuinely new treatments. Reaching novel territory is where the company places its bet.

The $2.1 Billion Round and the Novartis and Eli Lilly Deals

Major funding and pharma partnerships signal industry confidence in the AI-first approach.

›The $2.1 billion raise ranks among the largest biotech financings on record and was led by Thrive Capital, with backing from Alphabet, GV, and other investors.
›Eli Lilly paid $45 million upfront with up to $1.7 billion in milestone payments tied to results.
›Novartis paid $37.5 million upfront with up to $1.2 billion in potential milestones.
›The money funds the engine's growth, an internal drug pipeline, and hiring across research, engineering, drug design, and clinical work in London, Cambridge in Massachusetts, and Lausanne.

What Agentic AI Workflows Might Mean for Future Drug Discovery

Company leaders see automation extending beyond prediction into the full research loop.

Adrian Stecuła, a group leader in the company's machine learning organization, describes a future built on what he calls agentic workflows. In that setup, AI systems would generate hypotheses, test them, and analyze the results with far less manual effort at each step.

The idea points toward a research loop where the model proposes a new molecule, checks how it might behave, and learns from the outcome. Stecuła frames this kind of self-directed cycle as part of where the field is heading rather than a finished product available today.

A Plain-Language Reality Check on AI Drug Discovery

For all the progress, the people building these tools urge caution.

For more than a decade, AI has been promoted as a way to speed up drug discovery, yet few AI-designed medicines have reached patients despite billions invested. Part of the reason is that careful drug testing cannot be rushed, and part is that drug development is simply hard.

Stecuła warns against assuming that accurate structure modeling means the whole problem is solved. Knowing a protein's shape and even its hidden pockets is one early piece of a long chain that includes safety testing, manufacturing, and years of clinical trials. The cereblon result is encouraging evidence, not a finished cure.

Frequently Asked Questions

What is Isomorphic Labs?

Isomorphic Labs is a drug-discovery company spun out of Google DeepMind. It builds on DeepMind's AlphaFold protein-folding research and uses AI to design new medicines, working with partners including Novartis and Eli Lilly.

What does the Isomorphic Drug Design Engine (IsoDDE) do?

IsoDDE is a single AI system that predicts a protein's structure, identifies the pockets where drug molecules can bind, and estimates how tightly a molecule sticks to its target. These predictions help researchers find and design candidate drugs.

What is a cryptic pocket and why does it matter?

A cryptic pocket is a hidden cavity on a protein that looks absent in the protein's resting state and opens only when the right molecule binds. Finding these hidden sites lets researchers target proteins in new ways that standard analysis would miss.

How much funding has Isomorphic Labs raised?

Isomorphic Labs raised $2.1 billion, one of the largest biotech funding rounds on record. It also holds partnerships with Eli Lilly and Novartis worth billions in potential milestone payments.

Is AI close to designing medicines on its own?

Not yet. Company leaders say accurate structure and pocket prediction is real progress, but drug development still requires extensive safety testing and years of clinical trials that cannot be compressed.

Isomorphic Labs shows AI moving from predicting protein shapes toward the harder work of finding new drug targets, including hidden pockets earlier tools missed. The funding and pharma partnerships signal momentum, though turning these predictions into approved medicines still takes years of testing.