RFD4-Proteína

Researchers can use it to design proteins and other biomolecular structures conditioned on specified molecular features.

Availability

Downloadable from Hugging Face; interactive Studios are linked

Pricing model

Not established

Deployment

Local inference with the RFD4 Python package; hosted Studios are available

Privacy

Not established

Licence

BSD 3-Clause for the RFD4 model and weights; optional third-party assets have separate terms

• The model generates protein sequences and all-atom structures together. ^[1]

  It generates protein sequences and all-atom structures together, conditioned on the molecular context you want to preserve: a binding target, an enzyme motif, a ligand, or selected structural features.

• It can condition generation on a binding target, enzyme motif, ligand, or selected structural features. ^[1]

  It generates protein sequences and all-atom structures together, conditioned on the molecular context you want to preserve: a binding target, an enzyme motif, a ligand, or selected structural features.

• The repository provides native weights and a Diffusers-format repository with the same weights. ^[1]
• The native runtime supports Linux x86_64 and Python 3.12. ^[1]

  | **Runtime** | RFD4 Python package; Linux x86_64, Python 3.12, supported NVIDIA GPU and CUDA 13-compatible driver |

• The current release pairs a flow model with a unified autoencoder for proteins, nucleic acids, and small molecules. ^[1]

  The current release's autoencoder is **unified**: one autoencoder covering proteins, nucleic acids and small molecules, with a 147-class residue vocabulary in place of the previous 33-class protein vocabulary, and a 32-dimensional latent in place of 16.

• The model and weights are licensed under BSD 3-Clause. ^[1]

  | **License** | [BSD 3-Clause](LICENSE) for the RFD4 model and weights; optional third-party assets have separate terms |

• Create expressive speech from a short voice sample Daily brief 2026-10-27