Use RFpeptides Online
Commercially Available RFpeptides No-Code Web Server
RFpeptides: Diffusion-Based De Novo Design of Macrocyclic Peptide Binders
RFpeptides is a deep learning pipeline for the de novo design of cyclic peptide (macrocycle) binders against protein targets. It extends the RoseTTAFold2 and RFdiffusion models with a cyclic relative position encoding so that generated backbones close into a ring, and it combines this with ProteinMPNN sequence design and structure-based filtering.
Macrocyclic peptides sit between small molecules and antibodies: they can engage protein surfaces with high specificity while remaining compact. Designing them has traditionally depended on known binding partners or extensive screening. According to the authors, RFpeptides does not require any known binding partners for the target, which makes it applicable to targets with little prior ligand information.
The workflow was experimentally tested on four targets, with designs of roughly 12 to 18 residues, and several designs were confirmed by X-ray crystallography.
Feature | Details |
|---|---|
Method type | Diffusion-based macrocycle backbone generation (RFdiffusion with cyclic position encoding) |
Sequence design | ProteinMPNN, with Rosetta Relax in iterative rounds |
Inputs | Target structure or sequence, optional hotspot residues, target macrocycle length |
Design size tested | About 12 to 18 residues |
Targets tested | MCL1, MDM2, GABARAP, RbtA |
How RFpeptides Designs Macrocyclic Binders
The pipeline follows a generate, design and filter pattern:
Backbone generation: RFdiffusion denoises randomized starting atoms stepwise, using a cyclic relative position encoding so the peptide backbone forms a closed macrocycle.
Sequence design: ProteinMPNN proposes sequences for each backbone, and Rosetta Relax steps are used in the iterative rounds.
Filtering: Candidates are ranked using AfCycDesign predictions (interface PAE and similarity to the design model), RoseTTAFold2 re-prediction (used for MCL1 only), and Rosetta metrics such as ddG, SAP and CMS.
Experimental testing: Selected designs were synthesized, cyclized and tested for binding.
Optional design controls
Users can supply hotspot residues to steer binding to a chosen site (hotspots were used for GABARAP and RbtA) and choose the macrocycle length.
Experimental Results Across Four Protein Targets
Target | Designs tested | Outcome |
|---|---|---|
MCL1 | 14 of 27 selected designs | Three bound; best affinity about 2 uM; crystal structure at 2.1 A with 0.7 A macrocycle C-alpha RMSD to the design |
MDM2 | 8 of 11 selected designs | Three bound; best affinity about 1.9 uM |
GABARAP | 6 designs synthesized | Best two bound with affinities of 6 nM and 36 nM |
RbtA | 11 designs synthesized | Four showed binding; best affinity 9.4 nM; complex crystal structure at 2.6 A with 1.4 A macrocycle RMSD |
The crystal structures closely matched the computational models for MCL1 and RbtA, supporting the accuracy of the design approach.
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 RFpeptides on Tamarind Bio
Inhibitors of protein-protein interactions: The authors designed macrocycles against MCL1 and MDM2, two well-studied interaction targets.
Binders for autophagy-related proteins: Nanomolar binders were obtained against GABARAP.
Binders for bacterial proteins: Designs against RbtA reached nanomolar affinity.
Targets without known ligands: Because no known binding partner is required, the method can start from a structure alone.
How to Use RFpeptides on Tamarind Bio
Open the tool: Log in to tamarind.bio and open RFpeptides.
Provide the target: Supply the target protein structure or sequence, as the method defines (input format as exposed in the interface).
Choose hotspots: If exposed, specify hotspot residues to direct binding to the site of interest.
Set macrocycle length: Select the target ring size; the paper tested about 12 to 18 residues.
Run the job: Submit the run and generate multiple candidate designs.
Download and rank: Download the designs and rank them using interface PAE and the structure-based metrics described above.
Move to the lab: Synthesize top candidates and test binding experimentally.
Things to Keep in Mind
Hit rates were modest for MCL1 and MDM2, with three binders in each tested set.
Several designs failed at synthesis: cyclization yields were too low for 13 MCL1 designs and 3 MDM2 designs.
The GAB_D8 crystal structure showed two bound conformations, one partly deviating from the design model, and GAB_D23 deviated in one turn region.
Non-canonical amino acids, crosslinkers and other cyclization chemistries are future work, not part of the reported results.
Source: bioRxiv preprint, DOI 10.1101/2024.11.18.622547, https://www.biorxiv.org/content/10.1101/2024.11.18.622547v1