Google DeepMind and EMBL-EBI add millions of protein complexes to AlphaFold Database

Google DeepMind said on X on March 17, 2026 that it is expanding AlphaFold Database with millions of AI-predicted protein complex structures in collaboration with EMBL-EBI, NVIDIA, and Seoul National University. In the linked EMBL announcement, the partners say the release is the largest dataset of protein complex predictions currently available to the public.

The scale is substantial. Google DeepMind said AlphaFold Database has already been used by more than 3.3 million researchers worldwide, while EMBL notes the resource has served more than 3.4 million users across 190 countries. The collaboration says it has already calculated predictions for 30 million protein complexes, with the first public tranche centered on 1.7 million high-confidence homodimers and another 18 million lower-confidence homodimers expected through the FTP release path shortly after the announcement.

This matters because proteins rarely act alone. Much of biology depends on how proteins bind, assemble, and interact with each other. A database focused only on single-protein structures is extremely useful, but a resource that begins to expose large numbers of protein complexes can have broader value for disease research, drug discovery, and basic cell biology. The partners said they prioritized proteins linked to human health, including homodimers from 20 well-studied organisms and proteins relevant to the World Health Organization's bacterial priority pathogens list.

The infrastructure requirement also shows why an open release is significant. EMBL says recreating the full set would require roughly 17 million GPU hours, an amount of compute far beyond what most academic labs can access. Publishing the results lowers the barrier for researchers who want to explore interaction hypotheses without reproducing the entire pipeline themselves.

For the broader AI-for-science field, the update is another sign that the AlphaFold ecosystem is moving from a landmark demonstration to a continuously expanding knowledge platform. If researchers can reliably mine these predicted complexes alongside experimental data, AlphaFold Database could become even more central to the early stages of therapeutic discovery and molecular biology research.