RFD4-Proteina#

A next generation sequence structure codesign model for proteins, RNA, DNA and small molecule complexes.

Capable of satisfying robust conditioning and generating plausible sequences without further refinement.

Developed by the Baker Lab × NVIDIA GenAIR, with the Institute for Protein Design at the University of Washington.

We gratefully acknowledge the Jen-Hsun and Lori Huang Foundation for providing a computational grant used for this study.

NVIDIA NVIDIA Institute for Protein Design, University of Washington Institute for Protein Design, University of Washington RosettaCommons
A designed protein shown as a teal ribbon, wrapped around a small molecule ligand
              drawn in pink at its binding pocket.
Quickstart → Install, then generate your first designs on a GPU.

What you can design#

The released checkpoint supports all of these design tasks.

Task

Status

Unconditional generation

Supported

Motif scaffolding

Supported

Enzyme design

Supported, with a worked example

Small molecule binder design

Supported

Protein binder design

Supported

DNA binder design

Supported

State-of-the-art performance across relevant biomolecular tasks#

Average per-target success rate, unclustered
Grouped bar chart of average per-target success rate on three tasks. Enzyme
                design: RFD4-Proteina 0.36, Proteina-Complexa 0.11, RFD3 0.09. Small molecule
                binders: RFD4-Proteina 0.62, DISCO model 0.27, RFD3 0.24. Protein binder design:
                RFD4-Proteina 0.15, Proteina-Complexa 0.04, RFD3 0.04. Error bars on every bar.
Enzyme design Small molecule binders Protein binder design
Within each task, bars are RFD4-Proteina, the strongest open baseline, and RFD3.