Google DeepMind unveils an improved version of its AI model that predicts not only the structure of proteins, but also the structure of “all molecules of life.” The work of the new AlphaFold 3 model will help researchers in medicine, agriculture, materials science and drug development test potential discoveries.
Previous versions of AlphaFold only predicted the structures of proteins. AlphaFold 3 goes beyond this and can model DNA, RNA and smaller molecules called ligands, expanding the model’s capabilities for scientific use.
DeepMind says the new model has a 50 percent improvement in prediction accuracy compared to its previous models. “AlphaFold 2 was a huge milestone in structural biology and opened up all kinds of amazing research,” DeepMind CEO Demis Hassabis told reporters in a briefing. “AlphaFold 3 is a step towards using AI to understand and model biology.”
AlphaFold 3 has a library of molecular structures. Researchers enter a list of molecules they want to combine. AlphaFold 3 then creates a 3D model of the new structure using a diffusion method. Diffusion is the same type of AI system that AI image generators like Stable Diffusion use to compose photos.
According to DeepMind, Isomorphic Labs, a drug discovery company founded by Hassabis, uses AlphaFold 3 for internal projects. So far, the model has helped Isomorphic Labs improve its understanding of new disease targets.
In parallel to the model, DeepMind also makes the AlphaFold Server research platform available to some researchers free of charge. The server powered by AlphaFold 3 allows scientists to make biomolecular structure predictions regardless of whether they have computing power. Hassabis said the server is available for academic, non-commercial use, but Isomorphic Labs is working with pharmaceutical partners to use AlphaFold models for drug discovery programs.
Google says it is working with the scientific community and policymakers to use the model responsibly. Google says in a paper that some biosecurity experts believe AI models “could lower the hurdle for threat actors and allow them to collaborate with other technologies to design and develop pathogens and toxins that are more transmissible or harmful.”
The company said it worked with industry experts, biosecurity, research and industry specialists to understand the risks associated with AlphaFold 3 prior to its launch.