Isomorphic Labs released technical details about its Drug Design Engine (IsoDDE) in February, and the scientific community’s reaction was a mix of awe and frustration. The system appears to represent a major leap forward in AI-driven drug discovery - one that researchers cannot access or replicate.
What IsoDDE Claims to Do
IsoDDE is a unified AI system that handles multiple drug design tasks that previously required separate tools: predicting protein structures, determining how drugs bind to targets, estimating binding strength, and discovering previously unknown binding pockets.
The headline claims are significant:
- Doubles AlphaFold 3’s accuracy on the most challenging protein-ligand structure predictions (novel pockets and ligands dissimilar to training data)
- Outperforms AlphaFold 3 by 2.3x and the open-source Boltz-2 by 19.8x on antibody-antigen predictions in the high-accuracy regime
- Surpasses physics-based methods like FEP for binding affinity prediction while running in seconds rather than hours
- Discovers hidden binding pockets using only amino acid sequences, approaching the performance of experimental techniques like fragment-soaking
The system can model complex biological behaviors like induced fits (proteins reshaping to accommodate drug molecules) and cryptic pocket opening (hidden binding sites that only appear during molecular interactions).
The Scientific Reaction
Mohammed AlQuraishi, a computational biologist at Columbia University who works on open-source AlphaFold variants, called IsoDDE “a major advance, on the scale of an AlphaFold 4.” He was particularly struck by the system’s ability to generalize to molecules vastly different from its training data - “the really hard problem” in computational drug design.
But AlQuraishi’s next statement captured the scientific community’s dilemma: “The problem, of course, is that we know nothing of the details.”
Diego del Alamo at Takeda Pharmaceuticals raised questions about the source of IsoDDE’s performance gains. Isomorphic Labs has access to proprietary drug development data that academic researchers cannot obtain. “We don’t know how impactful that extra data is,” del Alamo noted.
Gabriele Corso from the Boltz non-profit, which develops open-source alternatives to AlphaFold, tried to find a silver lining: IsoDDE represents “a new baseline to match - but also to pass.”
The Open Source Problem
When DeepMind released AlphaFold 2 in 2020, it made the system freely available and published detailed papers enabling others to understand and build upon the work. This openness sparked an explosion of research and led to open alternatives like OpenFold, ESMFold, and Boltz.
Isomorphic Labs chose a different path. The company published a 27-page technical report that describes what IsoDDE can do without explaining how it achieves those results. The system remains entirely proprietary.
This matters because drug discovery is expensive and slow. If AI can genuinely accelerate the process, the approach taken by the leading systems shapes who benefits. Open tools let researchers everywhere contribute to and build upon advances. Proprietary tools concentrate capability in well-funded labs and pharmaceutical partnerships.
Isomorphic Labs, spun out of DeepMind in 2021, exists to commercialize AI drug discovery. Its business model depends on maintaining competitive advantage. But the contrast with DeepMind’s earlier approach is stark.
What This Means
IsoDDE appears to represent a real technical advance. The benchmarks, if accurate, suggest meaningful progress on problems that have stymied drug discovery for decades.
But the advance comes with conditions. Researchers cannot verify the claims independently. They cannot build upon the system or adapt it to their specific problems. The technology exists, but for most of the scientific community, it might as well not.
The open-source community continues developing alternatives. Boltz-2 and other projects aim to match or exceed proprietary systems while remaining freely available. Whether they can close the gap IsoDDE has opened remains uncertain.
For now, the state of AI drug discovery is split: Isomorphic Labs claims to have solved hard problems that others are still working on, and nobody outside the company can check their work.