Nvidia and Google DeepMind release predicted protein structures for 2,800 viruses to open database

Nvidia, Google DeepMind, and the European Bioinformatics Institute (EMBL-EBI) have released predicted three-dimensional structures for protein complexes across more than 2,800 viruses. The data is now available to any researcher through the AlphaFold Database, marking the largest coordinated effort to date to open-source structural biology for pathogens.

The predictions were generated using AlphaFold 2, DeepMind's AI model for protein folding, adapted to run efficiently on Nvidia GPUs through the BioNeMo Inference Runtime. The combination made it possible to scale inference to thousands of viral proteomes and to predict not just individual proteins but complexes — groups of interacting proteins that are often the real targets for vaccines and drugs. By comparison, traditional experimental methods such as X-ray crystallography take years and thousands of dollars per structure; here the turnaround is minutes per structure, with batch processing.

According to project data, roughly 30% of the protein interactions added to the database have never been documented in the Protein Data Bank, the central repository for experimentally determined structures. That represents a window into biological mechanisms that have not yet been studied. The Center for Global Development estimates a roughly 50% probability of a pandemic on the scale of Covid-19 by 2050. Joe Grove, professor of molecular virology at the MRC-University of Glasgow Centre for Virus Research, says the goal is to "bank knowledge in advance" for when the next virus arrives without warning — unlike the head start science had with coronaviruses.

Risha Patel, head of life sciences partnerships at DeepMind, notes the ambition has always been to make fundamental biology accessible at scale. Chris Dallago, head of applied research for digital biology at Nvidia, describes the database as a "hypothesis-generation engine": researchers can now explore protein interactions as whole complexes rather than as single molecules. To enable independent follow-on work, Nvidia is simultaneously releasing the BioNeMo Structure Prediction Pipeline, a GPU-accelerated workflow that lets any lab go from protein sequence to predicted 3D structure on targets of its own choosing.

Most proteins do not act alone; they assemble into multi-molecular complexes to carry out complex functions, and those combined structures are often what a vaccine or drug must hit to disrupt viral activity. Understanding the structure of the SARS-CoV-2 spike protein was a cornerstone of vaccine design; for thousands of other viruses, no comparable structural knowledge exists today. The team systematically worked through virus families known to infect humans, from common cold viruses to emerging threats such as Mpox, and began filling that gap.