AlphaFold 3 Can Now Predict the Structure of Nearly All Molecular Types in the Protein Data Bank

AlphaFold 3 was unveiled in May 2024, a major evolution of the AI system that revolutionized structural biology. While its predecessor AlphaFold 2 could predict the 3D structure of individual proteins from their amino acid sequences, AlphaFold 3 goes dramatically further: it can predict the structure and interactions of nearly all molecular types present in the Protein Data Bank, including proteins, DNA, RNA, ligands, and small molecules used in drug development.

Published in the journal Nature, the research demonstrated that AlphaFold 3 achieves at least 50% better accuracy than existing methods for predicting how proteins interact with other molecules. For some categories, such as protein-ligand interactions crucial for drug design, the improvement is even greater. The system uses a novel diffusion-based architecture — similar in concept to the technology behind AI image generators — that assembles molecular structures atom by atom.

The achievement builds on the legacy of AlphaFold 2, which has already been used by over 1.8 million researchers since its public release in 2021 and has been credited with contributing to advances in malaria vaccines, cancer research, plastic-degrading enzymes, and antibiotic development.