NVIDIA and DeepMind open 2,800+ viral protein structures

NVIDIA, Google DeepMind and EMBL-EBI, joined by the Coalition for Epidemic Preparedness Innovations, Seoul National University, Sungkyunkwan University, the Swiss Institute of Bioinformatics and the University of Glasgow, released an open dataset of predicted 3D structures for the protein complexes of more than 2,800 viruses. The predictions were inferred with AlphaFold2 optimized through NVIDIA’s BioNeMo Inference Runtime and published through the AlphaFold Database.

What’s actually new

• Complexes, not just single proteins. Most proteins act in groups, and those assemblies are what a vaccine or drug has to disrupt, so the dataset covers interacting protein groups across each viral proteome rather than isolated chains.
• About 30% of the interactions are new to science, showing interaction shapes never documented in the Protein Data Bank, the main repository of experimentally determined structures.
• The generating pipeline is open too. NVIDIA is releasing the BioNeMo Structure Prediction Pipeline, the GPU-accelerated workflow that goes from protein sequence to predicted 3D structure, so researchers can run it on their own targets.
• Every prediction is labeled by confidence, and the release lands in a database that now holds more than 260 million protein and complex predictions.

Why it matters

The traditional route to a structure, crystallizing a protein and shooting X-rays at it, takes years and thousands of dollars per structure; AlphaFold2 cuts that to minutes and makes bulk prediction practical, so the bottleneck shifts to experimental validation of the high-confidence calls. The open pipeline matters as much as the data, because it lets labs generate structures for emerging targets instead of waiting for a curated release. The release also coincided with a UN General Assembly meeting on pandemic preparedness convened by the World Economic Forum, and about 30% novel interaction geometry is a meaningful expansion of what the field can even ask about.

(Original source: NVIDIA)