Researchers have released predicted three-dimensional protein complexes for more than 2,800 viruses through the open AlphaFold Database. The coalition includes Google DeepMind, Nvidia, EMBL-EBI, universities and epidemic-preparedness organizations, and focused on viral families known to infect humans.
The team used AlphaFold2 to infer how groups of interacting proteins may fit together, with Nvidia’s BioNeMo Inference Runtime helping scale the work across thousands of viral proteomes. Nvidia is also publishing the GPU-accelerated BioNeMo Structure Prediction Pipeline so laboratories can run a similar sequence-to-structure workflow on their own targets.
Protein complexes matter because many vaccines and drugs must disrupt interactions among several molecules, not a protein in isolation. The release covers familiar viruses as well as less-studied threats. The project says roughly 30% of the added interactions have shapes not previously documented in the Protein Data Bank, creating leads for further investigation.
These structures are predictions, not experimentally confirmed answers. Entries carry confidence labels, and scientists still need laboratory methods to validate important candidates. That limitation makes the database a hypothesis-generation resource rather than a substitute for experiments. It joins an AlphaFold collection that now contains more than 260 million predicted proteins and complexes.