Use RSO Binder Design Online
Commercially Available RSO Binder Design No-Code Web Server
RSO: Relaxed-Sequence Gradient Descent for De Novo Protein Design with AlphaFold2
RSO is a de novo protein design approach that hallucinates protein backbones by running gradient descent on a relaxed (continuous) sequence representation through a differentiable AlphaFold2. It is implemented in the ColabDesign framework and paired with ProteinMPNN for sequence design, then validated by re-prediction with AlphaFold2, ESMFold and OmegaFold.
Earlier protein hallucination methods often relied on Monte Carlo (MCMC) sampling over discrete sequences, which can require tens of thousands of iterations. The authors show that optimizing a relaxed sequence, where each position holds a 20-value vector that can take fractional, greater-than-one or negative values, converges much faster and yields confident, designable backbones.
Feature | Details |
|---|---|
Core idea | Gradient descent on relaxed sequence logits through AlphaFold2 |
Framework | ColabDesign |
Sequence design | ProteinMPNN |
Validation | Re-prediction with AlphaFold2, ESMFold and OmegaFold |
Scoring | pLDDT and interface PAE (iPAE) |
How Relaxed Sequence Optimization Hallucinates Protein Backbones
Backbone hallucination: Gradient descent updates a relaxed sequence so AlphaFold2 confidently predicts a structure, producing a new backbone.
Sequence design: ProteinMPNN designs sequences for the hallucinated or supplied target backbones.
Re-prediction: The designed sequences are folded again with AlphaFold2, ESMFold and OmegaFold to check that they recover the intended structure.
Evaluation: Predictions are scored by pLDDT and, for multimers, interface PAE.
Multimer design
For heterodimers, a 50-residue gap is placed in the input to AlphaFold2, and protomer chains are predicted separately and in all pairwise combinations.
Results: Faster Convergence and Experimental Validation
Measure | Result |
|---|---|
Convergence | Average half-life of about 20 iterations; typically under 100 iterations overall, versus 30,000 MCMC iterations for a cited luciferase design |
Backbone recovery | Median RMSD below 4 A for targets up to 600 amino acids |
Monomers expressed | 13 of 14 designs soluble with the correct molecular weight |
Thermal stability | CD spectra largely unchanged up to 95 C |
Heterodimers | 2 of 18 pairs tested formed heterodimers (C5-C6 and D9-D10) |
Cryo-EM | 5.7 to 5.9 A maps; fitted AF2 models within 0.36 to 0.544 A RMSD of refined models |
Other tested protocols, hard one-hot gradient descent and MCMC, did not reach pLDDT above 90 within the tested steps, and ESMFold reproduced the ProteinMPNN-redesigned backbones similarly to AlphaFold2 in single-sequence mode.
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 RSO on Tamarind Bio
Unconditional monomer design: Hallucinate new stable, well-folded proteins without a target structure.
Sequence redesign of target backbones: Use ProteinMPNN with RSO backbones for new sequences.
Oligomer design: The authors obtained homodimers, homotrimers and heterodimers.
Fast design iteration: Rapid convergence allows many candidates to be generated and filtered.
How to Use RSO on Tamarind Bio
Open the tool: Log in to tamarind.bio and open RSO Binder Design.
Define the design task: Choose the protein length, or provide a target or partner chain for multimer and binder-style designs, as exposed in the interface.
Set parameters: Adjust options such as iteration count if exposed; the paper reports typical convergence in under 100 iterations.
Run the job: Submit the run to hallucinate backbones and design sequences.
Download outputs: Download the designed structures and sequences.
Interpret confidence: Prioritize designs with high pLDDT and, for complexes, low interface PAE.
Cross-check: Re-predict with ESMFold or another model and take top designs to the lab.
Things to Keep in Mind
Backbone recovery efficiency drops for backbones larger than 600 amino acids.
Cryo-EM resolution was not high enough to build atomic models, so side-chain geometry could not be evaluated.
The heterodimer interaction tests were not exhaustive, and only 2 of 18 pairs formed heterodimers.
Source: bioRxiv preprint, DOI 10.1101/2023.02.24.529906, https://www.biorxiv.org/content/10.1101/2023.02.24.529906v1