AlphaFold Database Adds Millions of Complexes — Interaction Scale

Single-chain AlphaFold changed structure prediction. The proprietary jump here is proteome-scale complexes in the public database — millions of interaction models prioritized for human health, not another monomer dump.

A collaboration of EMBL-EBI, Google DeepMind, NVIDIA, and Seoul National University added millions of AI-predicted protein complex structures to the AlphaFold Database, including high-confidence predictions for homodimers and other complexes. Dataset priorities: proteins important for human health and disease; 1.7 million high-confidence homodimer predictions added, with more planned. Large-scale study covers over 31 million predicted homo- and heteromeric complexes; open release for systems biology and drug discovery. Early applications framed as hypothesis generation on cellular machinery and potential drug targets. Nature and bioRxiv coverage in mid/late March describe the “next level” pairing expansion; DeepMind/Isomorphic updates on AlphaFold 3 and complexes provide lineage context.

Claims vs checks

Database addition and counts are EMBL-EBI / collaborator primary. “High-confidence” is model-confidence filtering, not crystallographic confirmation. Nature/bioRxiv pieces are secondary scientific coverage of the same expansion.

Limits

• Predicted interfaces can err; experimental validation remains required for drug claims.
• Coverage skewed toward prioritized proteomes.
• File date (March 12) precedes some March 17–27 coverage — treat as the expansion window, not a single URL day.

Sources

• EMBL: “Millions of protein complexes added to AlphaFold Database” (March 18, 2026 coverage of the expansion).
• Nature: “AlphaFold database hits ‘next level’: the AI system now includes protein pairing” (March 17, 2026).
• bioRxiv: “AlphaFold Database expands to proteome-scale prediction of protein complexes” (March 27, 2026 preprint; March work).
• Google DeepMind and Isomorphic Labs updates on AlphaFold 3 and complexes (2026).