Use RFdiffusion All Atom Online
Commercially Available RFdiffusion All Atom No-Code Web Server
RFdiffusion All-Atom: De Novo Design of Proteins that Bind Small Molecules
RFdiffusion All-Atom (RFdiffusionAA) is a diffusion model that generates protein structures directly around small molecules and other non-protein molecules, building binding pockets from random distributions of residues. It was developed by Rohith Krishna, Jue Wang, Woody Ahern and colleagues in David Baker's group at the University of Washington Institute for Protein Design, and is created by fine-tuning RoseTTAFold All-Atom (RFAA) on diffusive denoising tasks.
Designing small-molecule binders is a grand challenge. Earlier efforts docked molecules into large sets of native or expert-curated scaffolds, and the original RFdiffusion used a heuristic attractive-repulsive potential that could not model the details of protein-small molecule interactions; none of its designs were experimentally validated. A method that builds proteins around the ligand to maximize favorable interactions could be broadly useful.
RFdiffusionAA is explicitly conditional: during training, substructures of native complexes (motifs) are provided as context, and in design the ligand conformation is always part of the motif. It can also scaffold discontiguous protein motifs with or without a small molecule.
Property | Detail |
|---|---|
Base model | RoseTTAFold All-Atom weights fine-tuned as a denoising diffusion model |
Conditioning | Small molecule conformation, optional protein motif |
Sequence design | LigandMPNN, an atom-context-aware ProteinMPNN update |
Computational scoring | Rosetta GALigandDock energies and AlphaFold2 consistency |
Experimentally tested targets | Digoxigenin, heme and bilin (phycoerythrobilin) |
How RFdiffusionAA Designs Ligand-Binding Proteins
Fix the ligand: A random Gaussian distribution of residue frames is initialized around a fixed small molecule.
Denoise: At each step the model predicts the fully denoised state and moves coordinates and orientations toward it while adding noise to match the next step.
Optional potentials: Auxiliary contact potentials were generally unnecessary but can tighten interfaces for larger ligands such as FAD.
Assign sequences: LigandMPNN designs sequences for each backbone.
Filter: Designs are scored with Rosetta GALigandDock and AlphaFold2 predictions before testing.
RFdiffusionAA Results: In Silico and Experimental Validation
Evaluation | Result |
|---|---|
Interfaces with negative Rosetta binding energy | 0% for protein-only RFdiffusion with a heuristic potential; 52% for RFdiffusionAA; 72% with an auxiliary contact potential |
AlphaFold2 consistency | At least 45% of structures with backbone RMSD under 2 Angstrom in all four small-molecule cases |
Digoxigenin | 4,416 designs screened in yeast; tightest binder 10 nM Kd, stable up to 95 C |
Heme | 168 designs tested; 135 expressed well, 96 showed spectra consistent with cysteine-bound heme, 38 of 45 purified were monomeric and retained heme |
Bilin | 94 designs; 9 binders found (9.6% hit rate); absorption maxima of 557, 605 and 607 nm vs 573 nm for the control |
The designed binders were dissimilar to natural proteins: for digoxigenin the most similar PDB protein binding a related molecule had a TM-score of 0.59, and no detectable sequence similarity to any known protein was found. Most heme binders retained heme binding above 85 C.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with RFdiffusionAA on Tamarind Bio
Therapeutic-binding proteins: Design binders for small molecules such as digoxigenin, a cardiac disease therapeutic, which could help reduce toxicity.
Enzyme platforms: Build heme-binding scaffolds with an open axial site as a starting point for new enzymes.
Light-harvesting proteins: Tune the spectra of bilin-binding proteins for antenna complexes or fluorescent probes.
Motif scaffolding: Include a specific protein functional motif and design the surrounding structure.
How to Use RFdiffusionAA on Tamarind Bio
Open the tool: Log in at tamarind.bio and open RFdiffusion All-Atom.
Provide the ligand: Supply the small molecule target and its conformation, as the tool allows.
Define the design: Set the protein size or an optional motif to keep, if exposed.
Set parameters: Choose the number of designs and any contact potentials, if exposed.
Run: Launch the diffusion job.
Download backbones: Retrieve the generated protein structures.
Design sequences: Use a ligand-aware tool such as LigandMPNN, as in the paper.
Filter and test: Rank by AlphaFold2 consistency and Rosetta interface energy, then validate experimentally.
Things to Keep in Mind
Only a fraction of designs succeed, so many designs must be generated and filtered.
The experimental validation covers three targets: digoxigenin, heme and bilin.
Success depends on the ligand conformation supplied, which is always included as context.
The authors note that RFAA accuracy needs to improve further to have a big impact on drug discovery.
Source: Krishna R, Wang J, Ahern W, Sturmfels P, Venkatesh P, Kalvet I, Lee GR, Morey-Burrows FS, Anishchenko I, Humphreys IR, McHugh R, Vafeados D, Li X, Sutherland GA, Hitchcock A, Hunter CN, Baek M, DiMaio F, Baker D. Generalized Biomolecular Modeling and Design with RoseTTAFold All-Atom. bioRxiv preprint, posted October 9, 2023 (not peer reviewed in this version). DOI: 10.1101/2023.10.09.561603.