NetGoodIndexSubmit a correction

Benefit Ledger · verified · Biology

AlphaFold 3 predicts structures of proteins bound to DNA, RNA, and ligands

Google DeepMind and Isomorphic Labs published AlphaFold 3 in Nature, extending structure prediction from proteins to complexes that include nucleic acids, small molecules, ions, and modified residues.

8 May 2024Tier 4 HistoricMethodology 0.1

Current score

+10.77

30 base · Historic (tier 4 of 5, 30 pts)
× 0.9500 attribution · Primary causal contribution
× 0.7000 evidence · Peer review or independent validation
× 0.6000 realization · Experimentally validated
× 0.9000 durability
Event-level product before credit split: 10.77

Tier 4 reflects a field-level advance with a peer-reviewed method and public research access, not a civilizational health outcome. Attribution is high because the model is the result. Evidence is peer-review level (0.70), not independent wet-lab replication of every claim. Realization is experimental/tool deployment, not approved therapeutics. Durability is high for a reusable method.

What happened

On 8 May 2024, Nature published AlphaFold 3, a diffusion-based model that predicts the joint structure of biomolecular complexes across nearly all molecular types in the Protein Data Bank. The paper reported higher accuracy than specialized docking and nucleic-acid tools on several interaction classes, including protein–ligand and antibody–antigen cases relative to AlphaFold-Multimer v2.3. Google later released academic-use code and weights and a public AlphaFold Server. The result is a widely used research tool rather than an approved clinical product; several later open models (Chai-1, Boltz-1) independently targeted the same task.

Model attribution

AlphaFold

Produced the structure-prediction model jointly with Isomorphic Labs.

The published architecture and trained model are the central scientific artifact.

Attribution 0.9500 · Credit share 100% · Google DeepMind

Claims

  • AlphaFold 3 predicts joint structures of proteins with nucleic acids, ligands, ions, and modified residues in a single model.

    outcome · supported

  • The model improved accuracy over specialized tools on several interaction classes reported in the Nature paper.

    significance · supported

  • AlphaFold 3 has produced approved medicines.

    outcome · disputed

Sources

primary sources

Secondary domains: Medicine & Health, Chemistry

Revision history

  • 13 Sep 2026 · 0.00 10.77

    Initial adjudicated seed score under methodology 0.1.