AlphaFold won half a Nobel for predicting protein shapes; its free database holds 261 million
Two Google DeepMind researchers shared the 2024 chemistry Nobel for AlphaFold, which predicts how proteins fold. Its results are free.
Proteins are the machines that run living things, and what each one does depends on the shape it folds into. On 9 October 2024 the Royal Swedish Academy of Sciences gave half the Nobel Prize in Chemistry to Demis Hassabis and John Jumper of Google DeepMind for predicting those shapes with AlphaFold. The other half went to David Baker of the University of Washington for designing new proteins by computer.
The Academy says AlphaFold2 can predict the shape of virtually all of the roughly 200 million proteins researchers have identified, and that more than two million people in 190 countries have used it. The predictions sit in a free database run by EMBL-EBI with Google DeepMind, which now lists 260,986,406 of them. In March 2026 it added 1.7 million confident predictions of proteins paired with a copy of themselves.
AlphaFold 3, published in Nature in May 2024, goes further and predicts proteins locked together with DNA, RNA and small molecules, which is how most drugs work. DeepMind claims at least a 50% improvement over previous methods for some of those pairings. Its code was released for academic use in November 2024.
What it means
A question that previously needed laboratory work to answer is now a free lookup for any researcher, anywhere, which is part of why the prize came so soon after the tool. That helps the early stages of research into drugs, enzymes and disease. It does not mean protein biology is solved: every entry is a prediction with a confidence score attached, and labs still have to check the ones that matter.
Caveats
- Every entry is a prediction, not a shape measured in a lab. Confidence varies along each protein, and stretches the model is unsure about are common.
- AlphaFold gives one frozen shape. It doesn’t show how a protein moves, and the main database entries are single proteins on their own.
- AlphaFold 3 makes mistakes, including mirror-image errors in 4.4% of cases and occasional invented regions.
- The two-million-users figure comes from DeepMind and the Academy and hasn’t been independently checked.
- The structure count changes over time; the database front page shows both “over 260 million” and an older “over 200 million”.