The AI Drug Discovery Revolution: Isomorphic Labs’ ‘AlphaFold 4’ and the Future of Medicine
Nearly two years after Google DeepMind’s AlphaFold3 shifted the paradigm for drug discovery, its spin-off, Isomorphic Labs, has unveiled an even more powerful AI model – IsoDDE. This proprietary ‘drug-discovery engine’ is generating significant buzz, but also raising questions about accessibility and the future of open-source AI in the pharmaceutical industry.
A Leap Beyond AlphaFold3: What Makes IsoDDE Different?
Isomorphic Labs released a 27-page technical report on February 10th detailing IsoDDE’s capabilities. The model demonstrates impressive accuracy in predicting how proteins interact with potential drugs and antibody structures. Mohammed AlQuraishi, a computational biologist at Columbia University, describes it as a “major advance, on the scale of an AlphaFold4,” highlighting the substantial leap in performance.
The Challenge for Open-Source Alternatives
Unlike AlphaFold2, which was made accessible to researchers and detailed in journal articles, IsoDDE remains proprietary. This creates a challenge for scientists working on open-source alternatives. AlQuraishi notes the difficulty in replicating such results without understanding the underlying details of the model. “The problem, of course, is that we know nothing of the details,” he says.
Beyond Protein Structure: Predicting Drug-Protein Interactions
AlphaFold3 expanded beyond simply predicting protein structures to modelling interactions with other molecules, including potential drugs. IsoDDE builds on this foundation, excelling at predicting the strength of drug-protein binding – a crucial factor in therapeutic development. It also demonstrates state-of-the-art performance in predicting antibody interactions with their targets, which are vital for therapies generating billions in annual sales.
How IsoDDE Outperforms Existing Models
IsoDDE reportedly outperforms both open-source models like Boltz-2 (developed by MIT scientists) and traditional physics-based methods in determining binding affinity. Its ability to accurately predict interactions with molecules significantly different from its training data is particularly noteworthy, suggesting a novel approach to AI model design.
Isomorphic Labs’ Strategy: Data, Compute, and Algorithms
Isomorphic Labs’ president, Max Jaderberg, attributes IsoDDE’s success to a combination of computational power, data, and algorithms. Even as the company remains tight-lipped about specifics, they emphasize a “comprehensive” data strategy incorporating publicly available data, synthetic data, and licensed sources. They have also secured multi-billion pound drug-development deals with major pharmaceutical companies like Johnson & Johnson, Eli Lilly, and Novartis.
The Future of AI-Driven Drug Discovery
The emergence of IsoDDE signals a potential shift in the drug discovery landscape. While open-source initiatives like Boltz-2 continue to develop strides, the proprietary nature of IsoDDE raises questions about equitable access to cutting-edge AI technology. Gabriele Corso, the lead developer of Boltz-2, believes further improvements are possible with existing data, suggesting the race to develop superior AI models is far from over.
FAQ
- What is IsoDDE? IsoDDE is Isomorphic Labs’ proprietary AI model designed to accelerate drug discovery by predicting protein interactions with potential drugs.
- How does IsoDDE compare to AlphaFold3? IsoDDE is described as a more powerful model than AlphaFold3, achieving greater accuracy in predicting drug-protein interactions.
- Is IsoDDE open-source? No, IsoDDE is a proprietary model, unlike earlier versions of AlphaFold.
- What are the implications of a proprietary model? It creates a challenge for researchers developing open-source alternatives and raises questions about accessibility.
Pro Tip: Maintain an eye on developments in both proprietary and open-source AI models for drug discovery. The competition between these approaches will likely drive innovation and accelerate the development of modern therapies.
Did you know? Antibody therapies account for tens of billions of pounds in annual sales, making accurate prediction of antibody-target interactions a high-value application for AI.
What are your thoughts on the future of AI in drug discovery? Share your comments below!
