Google DeepMind last week unveiled AlphaFold 3, an evolution of its AI model that is now able to predict the structure and interactions of “all the molecules of life”. This marks a significant advance in molecular biology and could fundamentally change scientific research and drug development.
Technological foundations and developments
AlphaFold 3 builds on the achievements of its predecessor AlphaFold 2, which was released in 2020 and represented a groundbreaking advance in protein structure prediction. The model has already supported millions of researchers worldwide in the development of vaccines against malaria, cancer treatments and the design of enzymes. According to Google, the new version offers a 50% improvement in accuracy compared to existing methods, thanks in part to the use of an advanced diffusion network that works in a similar way to AI-based image generators.
How does AlphaFold 3 work?
The model analyzes an input list of molecules – including biomolecules, proteins, DNA, RNA and ligands – and generates a 3D structure from it. In addition, Google DeepMind provides the AlphaFold Server, a free research tool that gives researchers around the world access to the advanced capabilities of the model, regardless of their technical equipment.
Use in the development of drugs
Isomorphic Labs, a subsidiary of Alphabet that uses artificial intelligence in the development of medicines, is already using AlphaFold 3 in collaboration with pharmaceutical companies. This illustrates the technology’s potential to accelerate the drug development process and deepen our understanding of biological processes.
Biosafety and ethical considerations
During the development of AlphaFold 3, experts were extensively consulted to evaluate potential risks to biosafety. The technology could expand the possibilities for actors to develop more dangerous pathogens and toxins. This requires careful monitoring and regulation.
Outlook and expectations
According to Demis Hassabis, CEO and co-founder of Google DeepMind, the first drugs developed by AI could undergo clinical trials in the next few years. The publication of AlphaFold 3 could therefore herald a new era in scientific research, with far-reaching implications for various medical and technological fields.
