> AlphaFold_

For fifty years, the protein folding problem had been one of biology's greatest unsolved challenges. Proteins, the molecular machines that perform virtually every function in living cells, fold from linear chains of amino acids into intricate three-dimensional structures that determine their function. Predicting a protein's 3D structure from its amino acid sequence alone, without expensive and time-consuming experimental methods like X-ray crystallography, had been attempted by generations of scientists. Every two years since 1994, the Critical Assessment of protein Structure prediction (CASP) competition benchmarked the field's progress. For decades, progress was incremental.

Then came CASP14, held in late 2020, and DeepMind's AlphaFold 2 shattered every expectation. On the competition's primary metric, the Global Distance Test (GDT), AlphaFold 2 achieved a median score of 92.4 out of 100, where scores above 90 are considered competitive with experimental methods. The second-best performer scored around 75. The gap was so enormous that the organizers declared the protein folding problem, at least in its computational prediction form, essentially solved. John Moult, the founder of CASP, called it "a stunning advance," noting that some of AlphaFold's predictions were indistinguishable from experimental structures.

AlphaFold 2's architecture was a masterpiece of engineering. It combined a novel attention-based neural network (the Evoformer) that processed both amino acid sequences and evolutionary information from related proteins, with a structure module that directly predicted 3D atomic coordinates. The system learned to reason about the spatial relationships between amino acid residues using multiple sequence alignments and pair representations. In July 2021, DeepMind published the full method and open-sourced the code. Then, in partnership with the European Bioinformatics Institute (EMBL-EBI), they released predicted structures for nearly every known protein, over 200 million structures, essentially giving the entire field of biology a comprehensive structural atlas for free.

The impact on science has been transformative. Drug discovery pipelines that once required months of experimental structure determination can now begin with an AlphaFold prediction generated in minutes. Researchers studying diseases from malaria to Parkinson's have used AlphaFold structures to identify potential drug targets. In October 2024, Demis Hassabis and John Jumper of DeepMind were awarded the Nobel Prize in Chemistry for their work on AlphaFold, sharing the prize with David Baker for his complementary work on computational protein design. It was a fitting capstone: an AI system trained on decades of painstaking experimental data had not only solved a fundamental scientific problem but had accelerated the pace of biological discovery in ways that will be felt for generations.