Researchers are getting a new open-source resource aimed at helping the world prepare for future pandemics before an outbreak begins.
NVIDIA, Google DeepMind, EMBL-EBI and other global research organizations have released predicted 3D structures for protein complexes from more than 2,800 viruses. The data is being made openly available through the AlphaFold Database, giving scientists around the world access to structural information that could help guide future research into diagnostics, vaccines and treatments.
The effort is designed to address one of the major advantages scientists had during the COVID-19 pandemic: years of previous coronavirus research that helped researchers rapidly understand important viral proteins.
For other viruses, that knowledge may not exist.
Thousands of Viral Protein Complexes Predicted With AI
The new dataset was generated using AlphaFold2, Google DeepMind’s AI system for predicting the three-dimensional structures of proteins. NVIDIA’s BioNeMo Inference Runtime was used to optimize the process and help scale predictions across thousands of viral proteomes.
Instead of examining proteins only as individual molecules, the project focuses on protein complexes — groups of interacting proteins that can play important roles in how viruses function.
According to NVIDIA, around 30% of the protein interactions in the new dataset are completely new to science, with interaction shapes that have not previously been documented in the Protein Data Bank.
Risha Patel, life sciences partnerships manager at Google DeepMind, said the goal of the AlphaFold Database is to make foundational biological information more widely accessible and that the new viral-complex data could give researchers additional insights for preparing for future outbreaks.
Building Knowledge Before the Next Pandemic
The researchers say preparation cannot begin only after a new pathogen emerges.
Joe Grove, professor of molecular virology at the Medical Research Council-University of Glasgow Centre for Virus Research and a collaborator on the project, said a future pandemic could involve a virus for which scientists lack the prior knowledge available during COVID-19.
The new dataset is intended to help build some of that knowledge in advance.
A Center for Global Development analysis cited in the project materials estimates roughly a 50% chance of the world experiencing a pandemic as severe as COVID-19 by 2050.
The researchers’ approach is therefore focused on creating a large structural reference that scientists can investigate before an unfamiliar virus becomes an urgent global health threat.
Why Protein Structures Matter
Proteins rarely operate completely on their own. Many biological processes depend on interactions between multiple proteins, and those interactions can become important targets when scientists develop medicines or vaccines.
During the COVID-19 pandemic, structural knowledge of the coronavirus spike protein played an important role in vaccine research.
For many other viruses, comparable structural information is still unavailable.
Traditional techniques for determining protein structures can require extensive laboratory work and significant amounts of time. AI-based prediction can provide researchers with candidate structures much faster, although predictions still require experimental validation where appropriate.
The new dataset therefore does not replace laboratory research. Instead, it gives researchers structural hypotheses that can help guide further investigation.
NVIDIA Is Also Releasing the Prediction Pipeline
Alongside the dataset, NVIDIA is openly releasing its BioNeMo Structure Prediction Pipeline.
The GPU-accelerated workflow is designed to allow researchers to take protein sequences and generate predicted 3D structures for their own targets.
Chris Dallago, applied research science team lead in digital biology at NVIDIA, described the database as an engine for generating new biological hypotheses by allowing scientists to study protein interactions as complexes rather than only as individual molecules.
The project covers viral families known to infect humans, ranging from common-cold viruses to emerging threats such as Mpox.
Open Access Extends the Resource Beyond Major Research Centers
The collaboration includes the Coalition for Epidemic Preparedness Innovations, EMBL-EBI, Google DeepMind, NVIDIA, Seoul National University, Sungkyunkwan University, the Swiss Institute of Bioinformatics and the University of Glasgow.
Making the predictions openly available is intended to lower barriers for researchers, including scientists working in regions with fewer resources and those responding directly to outbreaks.
Jo McEntyre, interim director of EMBL-EBI, said open access to the information is important for research into viral diagnostics, treatments and vaccines.
The new release also expands the AlphaFold Database, which the project says now contains more than 260 million protein and protein-complex predictions covering nearly every cataloged protein known to science.
The predictions are labeled according to confidence, giving researchers information about how much trust to place in individual structural models.
What Happens Next
The newly released viral protein-complex structures are primarily a starting point for further research.
Scientists can use the predictions to identify potentially important protein interactions, develop hypotheses and prioritize structures for experimental investigation. Researchers can also use NVIDIA’s openly released pipeline to generate predictions for other protein targets.
The broader goal is straightforward: build more biological knowledge before the next major outbreak arrives, rather than beginning from scratch once a new pathogen is already spreading.
