On May 8, 2024, Google DeepMind and Isomorphic Labs introduced AlphaFold 3, a model that predicts not just the shape of a single protein but the joint structure of complexes - proteins together with DNA, RNA, small-molecule ligands, ions and chemically modified residues. The accompanying paper, “Accurate structure prediction of biomolecular interactions with AlphaFold 3” by Josh Abramson and colleagues, was published in Nature.
The core architectural change is a switch to a diffusion-based module that predicts raw atom coordinates directly. This lets one model handle many kinds of molecule without the special-case engineering that earlier tools required for each interaction type. The paper reports far greater accuracy for protein-ligand interactions than state-of-the-art docking tools, higher accuracy for protein-nucleic-acid complexes than nucleic-acid-specific predictors, and improved antibody-antigen prediction over AlphaFold-Multimer v2.3. Google’s announcement states at least a 50 percent improvement over existing methods for protein interactions with other molecule types.
Modeling how molecules interact - not just how a single protein folds - is what matters for drug discovery, since most drugs work by binding to a target. That is why the launch paired DeepMind with Isomorphic Labs, its drug-design sister company, which has exclusive commercial access. For non-commercial research, DeepMind opened most of the model’s capabilities for free through the AlphaFold Server.
AlphaFold 3 closes the arc that ran from the AlphaFold 2 contest win (2020) through the giveaway of the AlphaFold database (2021) to the 2024 Nobel Prize in Chemistry recognizing the work. It moves the technology from describing biology’s parts to predicting how they fit together.