NVIDIA and Google DeepMind Map Protein Complexes Across 2,800 Viruses
NVIDIA and Google DeepMind release open AlphaFold structures for 2,800 viruses, accelerating vaccine research and preparedness for future global pandemics.
Summary
On September 24, 2026, NVIDIA joined Google DeepMind, the Coalition for Epidemic Preparedness Innovations, EMBL-EBI, Seoul National University, Sungkyunkwan University, the Swiss Institute of Bioinformatics and the University of Glasgow to release predicted 3D protein-complex structures for more than 2,800 viruses through the AlphaFold Database Pandemic Preparedness Portal. AlphaFold2, accelerated by NVIDIA BioNeMo Inference Runtime on NVIDIA GPUs, processed viral families known to infect humans, from common-cold viruses to Mpox. About 30% of the inferred interactions have no previously documented shape in the Protein Data Bank. Confidence labels identify predictions requiring experimental verification.
NVIDIA also released the GPU-accelerated BioNeMo Structure Prediction Pipeline, letting researchers convert their own protein sequences into predicted structures. Conventional crystallization and X-ray analysis can cost thousands of dollars and take years per structure, while the AI workflow takes minutes and runs in bulk. The AlphaFold Database now contains more than 260 million protein and complex predictions covering nearly every cataloged protein. The Center for Global Development estimates a roughly 50% chance of another pandemic as severe as COVID-19 by 2050. Google DeepMind’s Risha Patel, NVIDIA’s Chris Dallago, EMBL-EBI’s Jo McEntyre and University of Glasgow virologist Joe Grove emphasized broader access, hypothesis generation and advance knowledge, particularly for lesser-studied viruses and low-resource settings. The release coincided with a United Nations General Assembly meeting convened by the World Economic Forum in New York City on pandemic prevention, preparedness and response.
Positives
- More than 2,800 viruses gain openly accessible predictions of protein-complex structures through the AlphaFold Database.
- About 30% of the predicted interactions represent shapes not documented in the Protein Data Bank.
- NVIDIA’s open BioNeMo Structure Prediction Pipeline lets researchers generate predictions for their own protein targets.
- AlphaFold2 reduces structure prediction from years of laboratory work to minutes of GPU computation.
- More than 260 million predictions now make the AlphaFold Database a broad foundation for biological research.
- Open access lowers structural-biology barriers for scientists studying lesser-known viruses in low-resource settings.
Risks & concerns
- A roughly 50% chance remains that a pandemic as severe as COVID-19 will occur by 2050.
- Thousands of viruses still lack the experimentally established structural knowledge that supported rapid COVID-19 vaccine development.
- AI-generated structures are predictions, and even high-confidence results require experimental verification.
- Traditional crystallization and X-ray methods can take years and cost thousands of dollars for each structure.
- The 30% of interactions without documented precedents highlights substantial remaining gaps in viral biology.