For decades, working out the shape of a single protein could cost a scientist years of painstaking lab work — then an AI called AlphaFold learned to do it in minutes and went on to map nearly every protein known to science, some 200 million of them, giving the - Space Daily

What the report says

Space Daily reported that AlphaFold2, the DeepMind protein-structure prediction system, transformed a long-standing bottleneck in structural biology by predicting many protein shapes far faster than traditional laboratory methods. The article traces the breakthrough to the November 2020 CASP14 assessment, where AlphaFold2 sharply outperformed other approaches, including in difficult cases without close known structural templates. The system was later described in Nature in July 2021 in a paper led by John Jumper.

The report explains why the advance mattered: experimental techniques such as X-ray crystallography, nuclear magnetic resonance and cryo-electron microscopy can take months or years for a single protein, and some proteins are especially hard to study. Space Daily cites the historical gap between roughly 100,000 experimentally determined unique structures at the time of AlphaFold2’s paper and billions of known protein sequences.

DeepMind and EMBL’s European Bioinformatics Institute launched the AlphaFold Protein Structure Database in July 2021 with about 350,000 predicted structures, then expanded it in July 2022 to more than 200 million, covering much of UniProt. The article says a 2024 database paper put the total at 214 million, with open access under a CC-BY-4.0 licence. It also notes that a July 2022 numerical error affecting about four percent of entries was corrected that November.

Space Daily emphasizes that AlphaFold predictions are not the same as experimental proof. Each entry includes confidence scores, with some regions or proteins requiring caution, and moving protein conformations remain a harder problem. The article also contrasts the open AlphaFold2 database with AlphaFold3, released in 2024 under more restrictive terms before code was later made available for non-commercial use.