Room 5 · Transformers · 2020

AlphaFold Solves Protein Folding

AlphaFold Solves Protein Folding (DeepMind, 2020): AI as a tool for scientific discovery.

In the CASP14 contest, AlphaFold 2 predicted the three-dimensional shape of proteins from their amino acid sequence with accuracy close to that of laboratory experiments. It was a problem that had been open for about 50 years. In 2022 a database with more than 200 million predicted structures was published.

Proteins are chains of amino acids that fold into precise three-dimensional shapes, and their shape largely determines what they do in the body. For decades, scientists had known that in principle the sequence contains the information needed to predict the shape. Doing it in practice was one of biology's great open problems. Determining a structure experimentally, with X-ray crystallography or electron microscopes, can take months or years for a single protein.

Since 1994 a community experiment called CASP has tested prediction methods every two years. Organizers release sequences of proteins whose structures have been solved in labs but not yet published, and teams submit blind predictions.

At CASP14, in late 2020, DeepMind's AlphaFold 2 produced predictions whose accuracy was, for many proteins, comparable to experimental methods. The organizers said the problem of predicting the structure of a single protein chain had been largely solved. The system combined deep learning with ideas from biology and physics, including an attention-based architecture that reasoned about which amino acids are near each other in space and how evolution has kept related sequences together.

In July 2021 the team, led by John Jumper and Demis Hassabis, described the method in Nature and released its code. Together with the European Bioinformatics Institute, EMBL-EBI, DeepMind opened a free database of predicted structures. In 2022 it was expanded to more than 200 million proteins, nearly every catalogued protein known to science.

Researchers now use these predictions to study diseases, design enzymes and plan experiments. A later version, AlphaFold 3, presented in 2024, extended predictions to interactions between proteins and other molecules.

The work earned Hassabis and Jumper half of the 2024 Nobel Prize in Chemistry, in Room 6.

Ask your AI: why does a protein's shape matter so much for medicine? Ask for one concrete example.

Did you know? Demis Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry for this work, together with David Baker, honored for protein design.

Source: doi.org/10.1038/s41586-021-03819-2

See it in the museum →

← GPT-3 and the Scaling Laws · Learning from Human Preferences →

All of Room 5: Transformers

Cite this: Nazareno, I. (2026). AlphaFold Solves Protein Folding (2020). New Mind Museum. Retrieved November 17, 2026, from https://newmindmuseum.com/museum/alphafold-solves-protein-folding-2020/