AlphaFold Database Expands With 8,000+ Virus Protein Structures to Prepare for the Next Pandemic

AlphaFold Adds Virus Protein Structures in Major Pandemic Preparedness Push

The AlphaFold Protein Structure Database has launched a pandemic preparedness portal, adding more than 8,000 virus protein dimers from 23 virus families with human-infecting members1

1

. This expansion represents a proactive use of AI to map viruses before they become health crises, addressing a critical gap in global health security.

The initiative brings together Google DeepMind, NVIDIA, the European Molecular Biology Laboratory's European Bioinformatics Institute (EMBL-EBI), and multiple research organizations to create an open dataset of predicted 3D structures for protein complexes from over 2,800 viruses2

2

. The collaboration spans institutions including the Coalition for Epidemic Preparedness Innovations, Seoul National University, the Swiss Institute of Bioinformatics, and the University of Glasgow.

Why Virus Protein Structures Matter for Future Outbreaks

When COVID-19 emerged, scientists had a crucial advantage. Decades of research on coronaviruses meant they understood SARS-CoV-2's key proteins well enough to design vaccines in record time2

2

. The next pandemic may not offer the same head start. Joe Grove, professor of molecular virology at the University of Glasgow's Centre for Virus Research, explains the urgency: "When the next pandemic happens, there may be something that comes out of the blue, and we'll be lacking the knowledge we had for COVID. What we're trying to do is stockpile some of that knowledge ahead of time"2

2

.

Many viral proteins do not act individually—they work in concert with partners to perform functions essential for viral replication, host cell entry, and immune evasion1

1

. These protein complexes often serve as targets for drug and vaccine development. Understanding the 3D structure of COVID-19's spike protein, for example, proved foundational to vaccine design2

2

.

Technical Breakthrough: Mapping Viral Proteomes at Scale

Researchers analyzed sequences of 41,774 proteins from approximately 2,800 viruses, including pathogens causing mpox, measles, hepatitis B, Zika, and dengue1

1

. Using AlphaFold optimized with NVIDIA BioNeMo Inference Runtime, the team predicted structures for 40,746 homodimers and nearly 1.7 million heterodimers1

1

.

Of these predictions, 2,749 homodimers and 5,279 heterodimers met accuracy thresholds for inclusion in the AlphaFold Protein Structure Database, though all predictions were made publicly available1

1

. Traditional methods for determining protein structures can take years and cost thousands of dollars per structure, but AlphaFold2 running on NVIDIA GPUs predicts structures in minutes2

2

.

About 30% of the protein interactions added to the database represent completely new structural insights never documented in the Protein Data Bank, the main repository of experimentally determined protein structures2

2

3

3

. Chris Dallago, applied research science team lead in digital biology at NVIDIA, describes the resource as "an engine for hypothesis generation," enabling biologists to investigate protein interactions as complexes rather than single molecules2

2

.

Addressing Viruses as a Blind Spot in Structural Biology

Despite the AlphaFold database holding predicted structures for most known proteins and serving more than three million users, viruses have been a blind spot1

1

. High-quality entries were lacking for many individual proteins, particularly in flaviviruses like Zika and dengue.

This gap exists because when some viruses replicate, their RNA molecules are first translated into polyproteins, which are then cut into individual functional molecules1

1

. This process makes it difficult to determine from a virus's genetic sequence where a viral protein begins and ends, leading to incomplete structure predictions. Researchers at the Swiss Institute of Bioinformatics in Geneva addressed this by identifying accurate sequences for thousands of viral proteins cut from polyproteins1

1

.

Open Science Approach Democratizes Access

Risha Patel, life sciences partnerships manager at Google DeepMind, emphasizes the mission: "Our ambition with the AlphaFold Database has always been to democratize access to foundational biology at scale. This collaboration to bring thousands of viral complexes into the database will equip scientists around the world with insights they need to help prepare for future outbreaks"2

2

.

NVIDIA is openly releasing the BioNeMo Structure Prediction Pipeline, the GPU-accelerated workflow used to generate the dataset, allowing researchers to predict 3D structures for their own targets2

2

. The AlphaFold Protein Structure Database now holds more than 260 million protein and protein complex predictions, covering nearly every cataloged protein known to science2

2

.

What Comes Next and What to Watch

The current release focuses on dimers, but larger protein complexes remain a priority. The spike protein enabling SARS-CoV-2 and other coronaviruses to infect host cells comprises three identical proteins, as does HIV's envelope-entry protein1

1

. Sameer Velankar, a bioinformatician at EMBL-EBI, confirms that including trimer predictions is an obvious next step, though determining the composition of bigger complexes presents challenges1

1

.

The predictions currently lack sugar molecules that adorn many viral proteins and help them evade immune detection1

1

. An analysis by the Center for Global Development estimates roughly a 50% chance of the world facing a pandemic as severe as COVID-19 by 20502

2

, underscoring the importance of this computational foundation for virology research and global health security.