Artificial intelligence has spent over a decade promising to revolutionize drug discovery, yet relatively few AI-designed medicines have reached patients. The bottleneck isn't just the lengthy testing timelines required for safety; it's that predicting a protein's shape is only the first step in creating a drug. Now, a Google DeepMind spinoff called Isomorphic Labs is tackling the harder problem: figuring out where and how drugs actually bind to proteins, and whether those bindings will work in the human body.

Isomorphic Labs recently raised $2.1 billion in funding and published a technical report describing its Isomorphic Drug Design Engine (IsoDDE), a unified computational system designed to discover the "pockets" on proteins where drug molecules can attach and to predict how proteins and drug molecules interact. The company has already signed major partnerships with pharmaceutical giants Novartis and Eli Lilly, signaling serious confidence in the approach.

Why AlphaFold Wasn't Enough for Drug Design?

When DeepMind's AlphaFold2 won the Nobel Prize in Chemistry, it solved one of biology's grand challenges: predicting how a protein folds into its three-dimensional shape from its amino acid sequence. AlphaFold3 expanded this capability to model how proteins interact with other biomolecules, including nucleic acids, small molecule drugs, ions, and other proteins, all within a single framework.

But here's the catch: both models have a critical limitation for drug discovery. As proteins become more novel and distant from the training data, the models' accuracy declines. For drug designers, this is a serious problem because the most promising new medicines often target never-before-observed binding sites on proteins, which means the models are least reliable exactly where they're needed most.

"AlphaFold2 was eventually recognized with the Nobel Prize, because it arguably solved the problem of protein folding. But proteins don't exist in a vacuum, right? They interact with a wide variety of other types of biomolecules, which involves nucleic acids, small molecule ligands, ions, and other proteins. AlphaFold3 introduced a way to model the rest of these cellular biomolecules as part of a single framework," explained Adrian Stecuła, a group leader in the machine learning organization at Isomorphic Labs.

Adrian Stecuła, Group Leader in Machine Learning at Isomorphic Labs

What Makes IsoDDE Different?

The Isomorphic Drug Design Engine takes a more comprehensive approach than structure prediction alone. Rather than just predicting where a drug molecule will bind to a protein, IsoDDE predicts multiple critical properties simultaneously:

• Structure Prediction: How the drug molecule and protein physically fit together in three-dimensional space.
• Pocket Identification: Finding binding sites on proteins, including hidden or "cryptic" pockets that only appear when the right molecule approaches.
• Binding Affinity Prediction: Measuring how tightly a drug molecule will stick to its target protein, a key factor in drug effectiveness.

The system also generalizes beyond small molecules to antibodies, molecular glues, and peptides, expanding the toolkit for tackling diseases where traditional drug approaches have failed.

The Cryptic Pocket Breakthrough: A Real-World Test?

Isomorphic Labs validated IsoDDE using a striking example: a protein called cereblon and its recently discovered "cryptic pocket." Cereblon is a critical protein in the cell's protein degradation pathway, and researchers had identified a completely novel binding site on its surface in a Nature paper published in January 2026.

The test was straightforward but revealing. Could IsoDDE find this never-before-disclosed pocket using only the protein's amino acid sequence as input? Yes. Could it accurately predict how drug molecules would bind to both the known binding site and the newly discovered cryptic pocket? Yes again. The model placed both ligands in exactly the correct positions, matching the crystal structures published in the research paper.

"In January of this year a Nature paper published a completely novel, never-before-observed cryptic pocket on the surface of this protein. First we asked the question: Can IsoDDE find this pocket just using the protein sequence as input? And we were able to perfectly predict the location of this cryptic pocket. Again, note that this pocket had never been disclosed before," stated Adrian Stecuła.

Adrian Stecuła, Group Leader in Machine Learning at Isomorphic Labs

Why This Matters for Patients?

Many diseases have known associated proteins, but the challenge is that these proteins often lack obvious binding sites that drugs can target. This is why certain diseases remain intractable despite decades of research. IsoDDE changes the equation by making it possible to find druggable mechanisms on proteins that previously seemed undruggable.

The practical implication is significant: diseases that have been considered too difficult to treat with small molecule drugs may now become accessible targets. This expands the universe of treatable conditions and could accelerate the path from protein discovery to clinical treatment.

How IsoDDE Advances Beyond Structure Prediction

The key insight from Isomorphic Labs is that drug discovery requires far more than accurate structure prediction. A unified system must predict multiple interdependent properties simultaneously:

• Multi-Property Modeling: IsoDDE doesn't just predict structure; it models binding affinity, ligand interactions with other proteins in the body, and how different therapeutic modalities will behave.
• Generalization to Novel Targets: The system is designed to perform well on protein-drug combinations that are distant from its training data, addressing the limitation that plagued earlier AlphaFold models.
• Therapeutic Modality Flexibility: The approach works for small molecules, antibodies, peptides, and molecular glues, not just one drug class.

Isomorphic Labs continues to improve performance on the endpoints it has disclosed and is pushing forward on additional capabilities not yet revealed publicly.

What's the Realistic Timeline for AI-Designed Drugs?

Despite the excitement around AI in drug discovery, there's a persistent misconception that solving protein structure means drug discovery is solved. The reality is more nuanced. Isomorphic Labs' work demonstrates that AI can accelerate the discovery phase, but the regulatory and testing timelines that ensure drug safety and efficacy cannot be easily compressed.

The company's partnerships with Novartis and Eli Lilly suggest that major pharmaceutical players believe IsoDDE can meaningfully shorten the early-stage discovery process, even if the overall timeline from discovery to patient remains measured in years rather than months. For diseases where no good treatments exist, even a modest acceleration in finding viable drug targets represents genuine progress.

" }