Read on nature.comIsomorphic Labs Unveils Proprietary 'AlphaFold 4' AI for Drug Discovery
Isomorphic Labs, the biopharmaceutical spin-off from Google DeepMind, has announced a proprietary AI model, IsoDDE, that scientists are calling an 'AlphaFold 4' for drug discovery. Unlike the publicly accessible AlphaFold systems, this new tool is being kept in-house. The model reportedly achieves state-of-the-art performance in predicting drug-protein binding affinity and antibody interactions, representing a significant leap forward for computational drug development. However, its closed-source nature raises questions about transparency and scientific collaboration in the field.
The landscape of AI-powered drug discovery has entered a new, more secretive phase. Isomorphic Labs, the London-based biopharmaceutical spin-off from Google DeepMind, has announced a proprietary artificial intelligence model that scientists are already comparing to a hypothetical 'AlphaFold 4.' This new system, dubbed IsoDDE (Isomorphic Drug Discovery Engine), represents a major leap in predicting how potential drugs interact with proteins, but unlike its groundbreaking predecessors, it will not be shared with the broader scientific community.
Announced in a 27-page technical report, IsoDDE builds directly on the foundation laid by AlphaFold 3, which itself was a significant evolution from the Nobel Prize-winning AlphaFold 2. While AlphaFold 2 revolutionized biology by predicting static protein structures, AlphaFold 3 expanded this capability to model proteins interacting with other molecules, including potential therapeutic compounds. Isomorphic's new engine takes this several steps further, achieving what experts describe as a generational advance in precision and utility for drug development.
A Proprietary Powerhouse
The most striking aspect of IsoDDE's announcement is its exclusive, closed-source nature. This marks a stark departure from DeepMind's previous strategy with AlphaFold, which was made publicly accessible and described in detailed journal articles in Nature. The IsoDDE technical paper, while showcasing impressive results, offers minimal insight into the model's architecture or training methodology. This has created a mix of awe and frustration within the computational biology community.
"It's a major advance, on the scale of an AlphaFold 4," says Mohammed AlQuraishi, a computational biologist at Columbia University, as reported by Nature. "The problem, of course, is that we know nothing of the details." AlQuraishi is working to develop fully open-source versions of AlphaFold, highlighting the growing divide between proprietary corporate AI and open academic research.
Key Capabilities and Breakthroughs
According to its technical report, IsoDDE demonstrates several state-of-the-art capabilities critical for drug discovery. A primary function is the precise prediction of binding affinity—the strength with which a potential drug molecule attaches to a target protein. This property is crucial for determining a compound's efficacy and is traditionally predicted using computationally intensive, physics-based simulation methods.
Isomorphic claims its AI not only outperforms these traditional methods but also surpasses recent open-source models like Boltz-2, developed at MIT, which itself was a notable attempt to replicate AlphaFold 3's capabilities. Furthermore, IsoDDE shows exceptional proficiency in modeling antibody-antigen interactions. Antibodies form the basis of many blockbuster biologic drugs, and accurately predicting their structure and binding is a multi-billion dollar challenge for the pharmaceutical industry.
The Implications of a Closed Ecosystem
The decision to keep IsoDDE proprietary signals a strategic shift for DeepMind's life sciences ambitions. Isomorphic Labs, led by DeepMind co-founder Demis Hassabis, is positioning itself as a drug discovery company rather than a pure research entity. By retaining exclusive access to its most powerful tool, it aims to build a competitive moat in the race to develop new therapeutics.
However, this move raises significant questions for the future of scientific progress in the field. The open release of AlphaFold 2 catalyzed an explosion of research and innovation across global biology labs. A closed-source model of similar or greater power could concentrate advanced capabilities within a single corporate entity, potentially slowing broader innovation. It also makes independent verification of the reported results impossible, a cornerstone of the scientific method.
For the pharmaceutical industry, IsoDDE represents both a promise and a threat. The promise is of dramatically accelerated and more accurate drug discovery pipelines. The threat is the potential for a new, powerful gatekeeper to emerge in the drug development process. Whether other companies will seek to license this technology or race to build their own equivalent models remains to be seen.
The unveiling of IsoDDE is a pivotal moment. It demonstrates that AI's potential in biology is far from tapped out, with models achieving ever-greater predictive power. Yet, it also underscores a growing tension between open science and proprietary commercial advantage. As AI becomes increasingly central to fundamental scientific discovery, the community must grapple with how to foster both rapid innovation and equitable access to the tools driving it forward.
