Isomorphic Labs' New AI Tool Doubles AlphaFold's Drug Discovery Accuracy
# Isomorphic Labs' New AI Tool Doubles AlphaFold's Drug Discovery Accuracy *System marks a shift from predicting protein structures to actually designing drugs on a computer* --- When Demis Hassabis was eight years old, he was already a chess prodigy.
Isomorphic Labs' New AI Tool Doubles AlphaFold's Drug Discovery Accuracy
System marks a shift from predicting protein structures to actually designing drugs on a computer
When Demis Hassabis was eight years old, he was already a chess prodigy. By seventeen, he'd coded the bestselling simulation game Theme Park. Today, the former child genius is attempting something far more ambitious: solving drug discovery.
In February 2026, Isomorphic Labs — the Google DeepMind spin-off Hassabis founded specifically for pharmaceutical research — announced its Drug Design Engine (IsoDDE), a unified computational system that more than doubles the accuracy of AlphaFold 3 on the most challenging protein-ligand predictions.
"Understanding biomolecular structures alone was not sufficient for unlocking real-world drug discovery programs in silico," the company wrote in its announcement. "Progress in rational drug design requires highly accurate predictive models across an expansive range of biochemical properties and interactions."
Beyond Structure Prediction
AlphaFold 3, released in 2024, revolutionized biology by predicting how proteins interact with drugs, DNA, and RNA. But benchmarks revealed a critical gap: the system struggled to generalize to molecular structures unlike anything in its training data — precisely the unexplored regions where many drug discovery opportunities lie.
IsoDDE addresses this by adding two critical capabilities:
- Improved generalization: On the "Runs N' Poses" benchmark (designed to test novel pocket and ligand prediction), IsoDDE more than doubles AlphaFold 3's accuracy on the most difficult cases.
- Binding affinity prediction: The system predicts small molecule binding affinities with accuracy exceeding gold-standard physics-based methods — at a fraction of the time and cost.
The Hassabis Thread
The announcement underscores a throughline in Hassabis's career: building AI systems that master games, then applying them to science. After DeepMind's AlphaGo defeated a world champion at Go — a feat many thought was a decade away — Hassabis redirected the technology toward protein folding, solving a 50-year grand challenge in biology.
Now he's going further. Isomorphic has partnerships with pharmaceutical companies to apply IsoDDE to real-world drug design challenges. The company's goal: develop new treatments for patients, designed entirely on computers.
To date, AlphaFold has been used by over 3 million researchers in more than 190 countries. But this next chapter isn't just about prediction — it's about creation.
Key facts:
Sources: Isomorphic Labs technical report, Nature (AlphaFold 3), Google DeepMind blog, company announcements.
