AI-predicted protein structures for more than 2,800 viruses go public · Digg

AI-predicted protein structures for more than 2,800 viruses go public

Nvidia, Google DeepMind, EMBL-EBI and research partners released the dataset to help scientists study viral interactions before future outbreaks.

1 Source, 8h ago, first seen 8h ago

TLDR

Nvidia and a research coalition have released AI-predicted 3D structures for protein complexes across more than 2,800 viruses through the AlphaFold Database. The partners used AlphaFold2 with Nvidia's BioNeMo tooling to model how viral proteins may interact, giving researchers ready-made hypotheses for diagnostics, vaccines and drug studies. Nvidia says about 30% of the added interactions are new to science. The structures carry confidence labels and still require experimental validation; prediction models have not reached experimental accuracy and can retain biases.

Featured Source

Scientists can now explore AI-predicted 3D structures for protein complexes across more than 2,800 viruses through the AlphaFold Database, following a new release from Nvidia, Google DeepMind, EMBL-EBI and a broader research coalition.

Nvidia says the project used AlphaFold2 with its BioNeMo Inference Runtime to work through viral proteomes at scale. The result is a public set of models showing how groups of viral proteins may fit and interact, including viruses known to infect humans from common-cold families to Mpox.

Protein complexes matter because biological functions often depend on several molecules working together. Their shape can suggest where a drug, antibody or vaccine might interfere with a virus. Traditional experimental techniques can take years and substantial expense for each structure, while AI can produce a prediction in minutes and process many candidates in bulk.

A head start, not laboratory proof

Nvidia says roughly 30% of the added protein interactions show shapes not documented in the Protein Data Bank, the principal archive of experimentally determined structures. That gives researchers new hypotheses to test rather than settled biological findings.

The distinction is important. The structures are labeled with confidence scores, and scientists still need experimental work to validate the most promising predictions. A related research manuscript says complex-structure prediction has not reached experimental accuracy and retains biases even when it produces useful starting points.

The collaboration also released the GPU-accelerated BioNeMo Structure Prediction Pipeline so researchers can run predictions on their own targets. Participants include the Coalition for Epidemic Preparedness Innovations, Seoul National University, Sungkyunkwan University, the Swiss Institute of Bioinformatics and the University of Glasgow, alongside Nvidia, Google DeepMind and EMBL-EBI.

The viral collection sits within a much larger expansion of the AlphaFold Database, which now contains more than 260 million protein and protein-complex predictions. The value of the new subset is its focus: scientists studying outbreaks can begin with searchable structural hypotheses instead of waiting for every viral interaction to be mapped from scratch.

AI-predicted protein structures for more than 2,800 viruses go public

Nvidia, Google DeepMind, EMBL-EBI and research partners released the dataset to help scientists study viral interactions before future outbreaks.

1 Source, 8h ago, first seen 8h ago

TLDR

Featured Source

A head start, not laboratory proof

1 Source

@nvidiaWith @GoogleDeepMind, @emblebi, and research partners, we’re making AI-predicted protein complex structures for 2,800+ viruses openly available. This gives scientists a head start in preparing for potential outbreaks.

Today's Rank

—

Not ranked yet

Combined views

45.7K

1 Source, first seen 8h ago

425 likes

AI-predicted protein structures for more than 2,800 viruses go public

TLDR

A head start, not laboratory proof

AI-predicted protein structures for more than 2,800 viruses go public

TLDR

A head start, not laboratory proof

1 Source

Combined views

A head start, not laboratory proof

A head start, not laboratory proof

Combined views

Sentiment

Summary

Sentiment

Summary

Sentiment

Summary

1 Source

Sentiment

Summary