AlphaFoldAlphaFold

For half a century, biologists had wrestled with the protein folding problem: given the chain of amino acids that makes up a protein, could you predict the intricate three-dimensional shape it folds into? Shape determines function, so the answer mattered enormously for medicine and biology.

In 2020, DeepMind's AlphaFold 2 effectively cracked it. At CASP14, a biennial competition that pits prediction methods against experimentally determined structures, it achieved accuracy rivaling slow, expensive laboratory techniques.

AlphaFold combined attention-based deep learning with knowledge of evolution and physics. It studied how related proteins across species varied together, using those patterns as clues to which parts of the chain sit close together in the folded structure.

The impact was immediate and vast. DeepMind released predicted structures for hundreds of millions of proteins, nearly every one known to science, as a free public database now used by researchers worldwide.

In 2024, Demis Hassabis and John Jumper of DeepMind shared the Nobel Prize in Chemistry for the work, a rare case of an AI system directly earning science's highest honor.