Sign in

AlphaFold solves protein folding

2020Turning point

AI predicts protein structures, solving a 50-year problem; it wins the 2024 Nobel Prize in Chemistry.

The story

On November 30, 2020, organizers of a biennial competition called CASP (Critical Assessment of protein Structure Prediction) announced that AlphaFold 2, an AI system built by the London company DeepMind, had predicted the three-dimensional shapes of proteins with accuracy rivaling laboratory experiments. On a 100-point scale of similarity to the true structures, it scored a median of about 92. Competition co-founder John Moult said the problem of predicting a single protein's structure had largely been solved.

Proteins are the working machines of life: enzymes, antibodies, muscle fibers, hormones. Each is a chain of building blocks called amino acids, strung together in an order set by a gene, and the chain folds itself into a precise shape that determines what the protein does. In the 1960s Christian Anfinsen showed that the sequence alone contains the information for the shape, so in principle it could be predicted. In practice the number of possible folds is astronomical. Scientists instead determined shapes experimentally with methods like X-ray crystallography, which could take months or years per protein. CASP began in 1994 to test prediction methods fairly, with teams predicting structures that had been solved but not yet published.

DeepMind, founded by Demis Hassabis and bought by Google in 2014, had already shown the power of deep learning with AlphaGo. Its first AlphaFold won CASP in 2018 but was far from lab accuracy. For AlphaFold 2, a team led by John Jumper designed a network using attention, the mechanism behind Transformers, trained on about 170,000 structures from the Protein Data Bank. It compares the target with related protein sequences from many species, since amino acids that mutate together tend to touch in the folded shape, and repeatedly refines a 3D model.

In 2021 DeepMind published the method and released its code, and with the European Bioinformatics Institute it made predictions freely available. By 2022 the database held over 200 million structures, covering nearly every known protein. Hassabis and Jumper shared the 2024 Nobel Prize in Chemistry with David Baker, who designs new proteins by computer.

Why it mattered

  • Biologists gained predicted shapes for over 200 million proteins, saving years of lab work.
  • Drug discovery sped up, as researchers could study the shapes of disease proteins with no known structure.
  • AI-designed proteins and enzymes, from vaccines to plastic-eating molecules, became a fast-growing field.
  • The 2024 Nobel Prizes recognized AI for the first time, in both physics and chemistry.

Drafted by AI (Claude) from the sources above and checked against them by AI; no historian has reviewed this page. Spot a mistake? Suggest a correction. How pages are made

Community notes

Loading notes…

Learn more ranked by votes