RFDpoly for design of RNA, DNA, and biopolymers is now available on GitHub

RFDpoly is a generative diffusion framework for the design of de novo biopolymers such as DNA, RNA, proteins, and mixed assemblies. Building on RFdiffusion, it extends the method beyond proteins by enabling the generation of backbones containing sugar and phosphate moieties to allow for the design of DNA and RNA structures.

In the original RFdiffusion framework, nucleic acids could only be treated as fixed ligands around which proteins were denoised. RFDpoly overcomes this limitation through two key developments: an expanded training set that includes diverse biopolymers, and new denoising procedures that operate on rigid-body parameters and torsion angles predicted by RoseTTAFold. As a result, RFDpoly can now generate protein structures, nucleic acid structures, and mixed assemblies that contain both.

Computational Testing

Computational benchmarks showed that RFDpoly can generate “short” RNA structures (<120 bases) with folds resembling motifs found in the PDB. For longer RNAs, the model produced novel topologies that were still predicted to fold reliably by AlphaFold3.

The authors also tested RFDpoly-generated RNA backbones in the Eterna OpenKnot design challenges. In a majority of cases, at least one sequence produced by NA-MPNN on an RFDpoly backbone outperformed both the wild-type sequences and MPNN redesigns of the experimentally characterized reference structures.

Experimental Validation

To experimentally validate the approach, the team used RFDpoly to design large, monomeric RNA structures composed of nearly cyclic repeating subunits known as RNA pseudocycles. These structures are large enough for characterization by electron microscopy, but required a new secondary-structure–template protocol to efficiently search for compatible base-pairing patterns. Experimental characterization yielded two designs that matched the original design models, two that matched AlphaFold3 predictions, and several that diverged from both, offering insights into the limits and opportunities of designability in high-complexity RNA systems.

RFDpoly was also applied to the design of protein–DNA complexes using a flexible scaffolding approach. The study demonstrated that RFDpoly can generate non-ideal nucleic acid geometries and design novel DNA-binding proteins.

We are excited to add this tool to Rosetta Common’s arsenal of protein design methods. You can find the inference code on GitHub. If you have any questions please reach out to the Rosetta Commons development team (Hope Woods, Rachel Clune, Rocco Moretti, Sergey Lyskov) either via Slack or through the Contact Form.