Google DeepMind and Isomorphic Labs have introduced AlphaFold 3, a significant upgrade to their protein structure prediction model that now includes DNA, RNA, and ligands, potentially rtransforming drug discovery and genomics research.
Building on the success of AlphaFold 2, which has been used by over 2 million researchers since its 2020 launch, Google DeepMind has unveiled AlphaFold 3. This new model expands beyond protein structure prediction to include all of life's molecules, addressing the evolving needs of biological and chemical research.
Jonas Adler, a research scientist at Google DeepMind, explains that the expansion was necessary to keep pace with current research frontiers, which increasingly focus on interactions between proteins and other biomolecules. Adrian Stecula from Isomorphic Labs highlights the model's potential in rational drug design, noting its ability to investigate binding of novel small molecules to drug targets.
The development of AlphaFold 3 presented unique challenges due to the vast diversity of small molecules compared to the ordered structure of proteins. To address this, the team expanded their training data significantly and shifted to a diffusion-based generative model for the final structure generation, similar to advanced image generation models.
A key innovation in AlphaFold 3 is its ability to predict both ordered and disordered regions of proteins, achieved by incorporating learnings from AlphaFold 2 into the new model's training.
To make AlphaFold 3 accessible, Google DeepMind has launched AlphaFold Server, a free tool allowing researchers to generate molecular complexes from their own sequences without coding knowledge. Since its May launch, researchers have already used it to generate over 1 million structures.
It is fair to say that AlphaFold 3 represents a significant leap forward in molecular structure prediction, potentially accelerating advancements in drug discovery, genomics, and other fields of biological research. As researchers begin to explore its capabilities, the impact of this tool on scientific discovery and medical advancements could be profound.