The AlphaFold Protein Structure Database, managed by Google DeepMind, Nvidia, and EMBL-EBI, has been updated with approximately 1.7 to 1.8 million new high-confidence protein complex predictions. This release focuses on viral proteomes, aiming to assist researchers in pandemic preparedness and the development of antiviral drugs.
The update includes 5,279 high-confidence heterodimers and 2,749 homodimers. Additionally, a companion dataset known as Viro3D provides over 85,000 predicted structures across more than 4,400 viruses, representing a 30-fold increase in viral protein data compared to previous experimental findings.
The collaboration leverages Nvidia’s GPU infrastructure for large-scale computational predictions, while DeepMind provides the AlphaFold technology. The database remains free for global researchers. Since its 2021 launch, AlphaFold has expanded to cover roughly 200 million protein structures.
While the database serves as a tool for hypothesis generation, developers note that these are AI-predicted structures rather than experimentally verified ones. Researchers are advised to use the data as a starting point, as reliance on inaccurate predictions could impact drug development efficiency.
