Google 2015 logo (Wikimedia Commons, public domain)
The expanded AlphaFold database now includes roughly 1.8 million protein complex predictions aimed at accelerating pandemic preparedness and antiviral drug discovery
Think of protein structures as the blueprints of biology. If you want to build a drug that stops a virus, you first need to understand the shape of the lock you’re trying to jam. Nvidia, Google DeepMind, and the European Molecular Biology Laboratory’s European Bioinformatics Institute (EMBL-EBI) just handed researchers the keys to roughly 2,812 viral proteomes worth of those blueprints, all predicted by AI and all freely available.
The update to the AlphaFold Protein Structure Database, released in March 2026, adds approximately 1.7 to 1.8 million high-confidence protein complex predictions. The focus this time: viruses, and the goal of giving scientists a head start before the next pandemic arrives rather than after.
What’s actually in the database #
The numbers here are staggering when you consider the baseline. A companion dataset called Viro3D provides over 85,000 high-quality predicted structures spanning more than 4,400 viruses. That expands the representation of viral proteins in the database by roughly 30-fold compared to existing experimental data.
The release includes 5,279 high-confidence heterodimers, which are complexes formed by two different protein chains, and 2,749 high-confidence homodimers, complexes of two identical chains. Beyond the headline figures, tens of millions of additional structures are available for researchers to download from AlphaFold’s broader initiatives.
Homodimers make up the bulk of the complex predictions. That matters because many viral proteins function as pairs or groups, and understanding those assemblies is critical to identifying drug targets.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Why this matters for drug discovery #
The collaboration between Nvidia and Google DeepMind is worth noting for what each party brings to the table. DeepMind developed AlphaFold, which won the 2024 Nobel Prize in Chemistry for its founders. Nvidia provides the GPU computing infrastructure that makes predictions at this scale feasible. EMBL-EBI hosts and curates the database, ensuring it remains freely accessible to the global research community.
Since its first major release in 2021, AlphaFold has transformed structural biology from a field defined by painstaking experimental work into one increasingly powered by computational prediction. The original database covered around 200 million protein structures across nearly all known organisms. This latest update sharpens the focus on the viral world specifically.
The risk, as always with AI predictions, is overconfidence. These are predicted structures, not experimentally verified ones. High-confidence scores from AlphaFold are remarkably accurate in most cases, but they are not infallible, and drug development decisions based on incorrect structural predictions could waste time and resources. The dataset’s value is as a starting point and hypothesis generator, not as a substitute for experimental validation.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our