London-based Isomorphic Labs, a spin-off from Google DeepMind, has announced a significant leap forward in artificial intelligence for drug discovery with its latest model, IsoDDE. The proprietary system, detailed in a technical report released February 10th, is already generating excitement among scientists for its ability to accurately predict how proteins interact with potential therapeutic molecules – a crucial step in developing new medicines. Experts are describing the advance as comparable to the arrival of a future generation of DeepMind’s technology, informally dubbed ‘AlphaFold 4’.
Unlike its predecessors, including the groundbreaking AlphaFold2 and AlphaFold3, IsoDDE’s inner workings remain closely guarded. This secrecy has sparked both admiration and frustration within the scientific community, particularly among those working on open-source alternatives. While AlphaFold’s earlier iterations were openly shared, allowing for collaborative development and refinement, IsoDDE represents a shift towards a more commercially focused approach to AI in drug discovery.
The core capability of IsoDDE lies in its ability to predict binding affinity – the strength with which a potential drug molecule attaches to a target protein. According to Isomorphic Labs’ report, the model outperforms both existing physics-based methods and the open-source Boltz-2, developed by researchers at MIT, in this critical area. This improved accuracy extends to predicting interactions with antibodies, which are key components of therapies generating tens of billions of pounds in annual sales, according to the report.
Predicting the Unpredictable: IsoDDE’s Novel Approach
Mohammed AlQuraishi, a computational biologist at Columbia University, highlighted the model’s impressive ability to predict interactions with molecules significantly different from those used in its training data. “That’s the really hard problem, and suggests that they must’ve done something pretty novel,” AlQuraishi stated. This suggests IsoDDE has moved beyond simply recognizing patterns in known data and is capable of generalizing to new and unfamiliar chemical structures.
Isomorphic Labs president Max Jaderberg acknowledged the model’s complexity, describing the underlying technology as “profoundly different” from other efforts in the field. Whereas, the company has no immediate plans to reveal the specifics of its architecture or algorithms, citing a combination of computational power, data, and algorithmic innovation as key factors. Jaderberg expressed hope that the technical report will inspire further development in the broader AI drug discovery space.
Data Strategy and Industry Partnerships
The success of IsoDDE is likewise raising questions about the role of data. Diego del Alamo, a computational structural biologist at Takeda Pharmaceuticals, noted on X (formerly Twitter) that Isomorphic Labs’ extensive partnerships with pharmaceutical companies likely provided access to valuable private structural data, potentially contributing to the model’s performance. Isomorphic Labs has indeed forged collaborations with industry giants including Johnson & Johnson, Eli Lilly, and Novartis, and has its own internal pipeline of clinical trials underway.
Isomorphic’s data strategy, according to director of machine learning Michael Schaarschmidt, is “quite comprehensive,” incorporating publicly available data, synthetically generated training data, and licensed datasets. However, Gabriele Corso, co-developer of Boltz-2 and leader of the non-profit Boltz, believes that proprietary data isn’t necessarily the key differentiator. Corso argues that significant improvements can still be achieved using publicly available data, suggesting that IsoDDE’s success may lie in algorithmic advancements rather than exclusive access to information.
The Future of AI-Driven Drug Discovery
The emergence of IsoDDE marks a pivotal moment in the application of artificial intelligence to drug discovery. While the proprietary nature of the model creates a divide between those with access and those developing open-source alternatives, it also underscores the growing commercial potential of AI in the pharmaceutical industry. The competition spurred by IsoDDE’s advancements is likely to accelerate innovation across the field, potentially leading to faster and more efficient development of life-saving therapies.
Looking ahead, the focus will be on translating these AI-powered predictions into tangible clinical benefits. Isomorphic Labs is already working with partners to explore the application of IsoDDE to specific drug development programs, and the results of these collaborations will be closely watched by the scientific community. The ongoing evolution of AI models like IsoDDE promises to reshape the landscape of pharmaceutical research and development in the years to reach.
What are your thoughts on the balance between proprietary AI development and open-source collaboration in the pharmaceutical industry? Share your perspective in the comments below.
