The AlphaFold Protein Structure Database opens

DeepMind and the European Bioinformatics Institute (EMBL-EBI) launched the AlphaFold Protein Structure Database, publishing around 350,000 predicted protein structures — covering essentially the entire human proteome, the roughly 20,000 proteins expressed by the human genome, plus another twenty organisms significant to research, including E. coli, the fruit fly, the mouse and the malaria parasite. It followed by a week the publication of AlphaFold 2’s full method and code in Nature, seven months after the system’s CASP14 result had first shown its accuracy.

The database made structures freely and openly available to any researcher, without requiring either the computational resources to run AlphaFold or the months to years that determining a structure experimentally, by X-ray crystallography or cryo-electron microscopy, could take. DeepMind chief executive Demis Hassabis called it “the most significant contribution AI has made to advancing scientific knowledge to date”; EMBL-EBI director Ewan Birney said it would be “one of the most important datasets since the mapping of the Human Genome.” Both organisations said they intended to expand coverage to more than 100 million structures, spanning most of the UniProt reference database, in the following year.

Early applications cited by DeepMind and EMBL-EBI included work on neglected tropical diseases, antibiotic resistance and enzymes capable of breaking down plastic. Because the database put structural predictions directly into a searchable, citable public resource rather than leaving AlphaFold as a system only DeepMind could run, it accelerated the shift from AlphaFold as a demonstrated capability to AlphaFold as everyday laboratory infrastructure — a pattern DeepMind and EMBL-EBI repeated the following year with a large expansion of the database’s coverage.