RFdiffusion3: De Novo Design of All-atom Biomolecular Interactions

RFdiffusion3 (RFD3) is a revolutionary diffusion model that elevates protein design by explicitly modeling all polymer atoms, including backbones and sidechains, alongside non-protein components like small molecules and nucleic acids. This all-atom design capability provides unprecedented precision and control, making it the most effective platform for generating complex functional proteins, including enzyme design and DNA binding proteins.

Key Innovations in RFdiffusion3

RFD3 represents a major leap from prior residue-level diffusion methods (like RFdiffusion1 and RFdiffusion2) and simplifies challenges faced by other all-atom models like AlphaFold3 (AF3) by focusing on design over prediction.

• Explicit All-Atom Design: RFD3 diffuses every individual atom (backbone and sidechain) of the protein, enabling the direct modeling and control of precise atom-level constraints, which is impossible with residue-level methods.

• Unified Biomolecular Interaction: It is explicitly trained to generate protein structures in the context of ligands, nucleic acids (DNA/RNA), and other non-protein constellation of atoms, providing a general solution for designing complex interacting systems.

• Unprecedented Control: The model natively supports and simplifies the specification of atomic-level design constraints for precise control over:

  • Hydrogen Bonding: Direct specification of donor and acceptor atoms for productive hydrogen bonding interactions with the target molecule.

  • Solvent Accessibility: Conditioning on the Relative Solvent Accessible Surface Area (RASA) to control the burial depth of ligand atoms.

  • Symmetry-Constrained Diffusion: Enables the rapid generation of symmetric structures (e.g., D2, C3, C5, C7 oligomers) by supplying symmetric noise as input.

• Superior Computational Efficiency: By utilizing a lean architecture and sparse attention, RFD3 achieves an improved performance compared to prior approaches with one tenth the computational cost. It is approximately 10-fold faster than RFdiffusion2 for typical protein lengths.

Broad Applications in De Novo Design

RFD3 has demonstrated superior performance across a diverse range of design challenges, validated by rigorous in silico benchmarking and experimental testing.

Enzyme Design and Catalysis

RFD3 significantly simplifies and improves enzyme design by allowing active sites to be scaffolded around minimal atomic motifs.

• Scaffolding Atomic Motifs: The full atomistic representation simplifies generating scaffolds around minimal motifs, eliminating the complexity of the hybrid residue-atom approach used previously.

• Experimental Validation: RFD3 successfully generated highly active cysteine hydrolases with Kcat/Km=3557 M^-1\s^-1 for an esterase reaction, exceeding previous designs.

Protein-Nucleic Acid and Protein-Small Molecule

The model is the first to effectively co-design the protein and the conformation of flexible non-protein targets.

• DNA Binding Proteins: Successfully designed and experimentally characterized a DNA binding protein (DBRFD3) that bound the target DNA sequence with an EC50 of 5.89 +/- 2.15 μM.

• Joint Conformation Sampling: RFD3 enables the joint sampling of the protein structure with the conformation of the target ligand geometry (for small molecules) or DNA conformation, surpassing previous methods that required rigid inputs.

Protein-Protein Interaction

RFD3 significantly improves upon RFdiffusion1 for protein binder design, consistently generating more diverse and novel solutions against therapeutically relevant targets.

• Diversity and Novelty: It samples significantly more diverse (8.2 vs. 1.4 successful clusters on average) and more novel folds than RFdiffusion1 for targets like PD-L1 and Tie2.

What is Tamarind Bio?

Tamarind Bio is a pioneering no-code bioinformatics platform built to democratize access to powerful computational tools for life scientists and researchers. Recognizing that many cutting-edge machine learning models are often difficult to deploy and use, Tamarind provides an intuitive, web-based environment that completely abstracts away the complexities of high-performance computing, software dependencies, and command-line interfaces.

The platform is designed provide easy access to biologists, chemists, and other researchers who may not have a background in programming or cloud infrastructure but want to run experimental models with their data. Key features include a user-friendly graphical interface for setting up and launching experiments, a robust API for integration into existing research pipelines, and an automated system for managing and scaling computational resources. By handling the technical heavy lifting, Tamarind empowers researchers to concentrate on their scientific questions and accelerate the pace of discovery. The Tamarind team hold information/data security as a top priority, as detailed in our Trust Center & Terms of Service, ensuring your data is safe on the platform.

How to Use RFdiffusion3 on Tamarind Bio

To leverage RFD3 all-atom design capabilities, a researcher can follow this streamlined workflow on the Tamarind Bio platform:

• Access the Platform: Begin by logging in to the tamarind.bio website.

• Select RFdiffusion3: From the list of available generative models, choose the RFdiffusion3 tool.

• Input Design Constraints: Provide the specific high-level design conditions, such as:

  • Target molecule coordinates (ligand or DNA sequence and initial conformation).

  • Atomic Motif coordinates (e.g., a catalytic triad for enzyme design).

  • Optional: Hydrogen bond donor/acceptor atoms or RASA constraints.

• Run RFD3 Generation: The platform executes the RFD3 diffusion trajectory, generating the all-atom structure (backbone and sidechains).

• Analyze Leads: The output provides the final protein backbone and sidechain coordinates, which can then be paired with a sequence design tool (like ProteinMPNN or LigandMPNN) for final sequence fitting and predicted validation.

Source