From Gaming Theory to Protein Structure Prediction

In 1961 Chris Anfinsen demonstrated that the linear sequence of amino acids in the polypeptide chain of a protein contained all the information required to dictate its folding into a unique 3-dimensional structure. Since then, biochemists have been on a long quest to be able to predict the structure of a protein directly from its amino acid sequence.

The 2024 Nobel Chemistry prize recognised the pioneering work of Demis Hassabis and John Jumper who successfully used computing and artificial intelligence (AI) to predict the structure of almost all known proteins. This talk will trace how this evolved, from the first determination of the 3D structure of myoglobin and haemoglobin by Kendrew and Perutz in 1962.

Demis Hassabis, at 13 already a chess grandmaster and in his teens a programmer and a games developer, used his experience with AI and neuroscience to develop AI models for popular board games (his company Deep Mind was sold to Google in 2014). He then directed his attention from 2017 into solving the protein folding problem with the AI programme Alpha Fold, and together with John Jumper they fundamentally transformed this into Alpha Fold2. By 2020 AlphaFold2 performed almost as well as X-ray crystallogra­phy in predicting protein structure.

At the end of the 1990s, David Baker (who received half of the 2024 Chemistry Nobel Prize), began to develop computer software that could predict protein structures, called Rosetta. At the same time, the field of protein design – to create bespoke proteins with new functions - also took off, and Baker decided to use Rosetta to design a new protein, Top7, which, in 2023, was shown to have the structure that was predicted. Since 2020 he has used Alpha Fold2 to design a series of incredible new proteins.