AlphaFold mapped over 214 million proteins — conservation scientists are finally using it

For sixty years, deciphering a single protein’s three-dimensional structure required months of crystallography or electron microscopy. AlphaFold, developed by Google DeepMind in partnership with the EMBL-EBI, reduced that to seconds. The 2021 landmark paper by Jumper et al. in Nature was described as “a solution to a 50-year-old grand challenge in biology.” Since then, the open database has grown to over 214 million protein structure predictions, covering virtually all catalogued proteins across hundreds of thousands of species.

Conservation biology arrived late to this resource — and is finding it unexpectedly useful. European researchers used AlphaFold to decode Vitellogenin (Vg), an immunity protein central to honeybee health. Those structural insights are now guiding breeding programmes aimed at populations already classified as endangered. Across a wider range of species, researchers are mapping proteins in organisms for which no structural data previously existed — filling a gap in the molecular record of life.

Three million researchers across more than 190 countries draw on the database, which has been cited in over 35,000 papers.

One limit runs through all of this: AlphaFold predicts structure, not function. Knowing a molecule’s shape is not the same as understanding what it does — wet lab validation remains essential. And protein maps, however precise, do not substitute for the habitat protections and policy commitments that determine whether an endangered species ultimately survives.

Further reading: The AI That Counts Species: When Artificial Intelligence Serves the Living World