Use AfCycDesign Online

AfCycDesign: Cyclic Peptide Structure Prediction and De Novo Macrocycle Design with AlphaFold

AfCycDesign is a modified version of AlphaFold that can predict the structures of cyclic peptides (macrocycles) and design new cyclic peptide sequences. It was developed by researchers at the University of Washington Institute for Protein Design and Harvard University, and was implemented in the ColabDesign framework.

Cyclic peptides are attractive because of their stability and binding properties, but computational support for them has lagged behind that for proteins. According to the authors, there are few experimentally solved cyclic peptide structures, and existing methods for predicting and designing them are computationally expensive.

AfCycDesign addresses this by adding a cyclic positional encoding to AlphaFold, so the N- and C-termini are treated as sequence neighbors. This lets the network reason about head-to-tail cyclized backbones directly, and supports fixed-backbone sequence design as well as de novo hallucination.

How AfCycDesign Predicts and Designs Cyclic Peptides

The method changes the relative positional encoding with an N x N cyclic offset matrix that is added to the pairwise features. The model is therefore invariant to circular permutations of the sequence.

1. Structure prediction: input a single cyclic peptide sequence (optionally with an MSA) and receive a predicted structure with pLDDT and PAE confidence.

2. Fixed-backbone sequence redesign: starting from a cyclic backbone, the design loss minimizes the cross-entropy between the predicted distogram and the target backbone distogram, moving through a three-stage protocol from continuous logits to a one-hot sequence.

3. Hallucination: sequence and structure are sampled jointly from random initialization, optimizing a loss that combines 1 minus pLDDT, PAE and a contact term.

AfCycDesign Benchmark Results for Cyclic Peptides

Structure prediction accuracy

On 80 NMR-derived cyclic peptides from the PDB (fewer than 40 residues, canonical amino acids), the median pLDDT was 0.88 and the median RMSD to the native structure was 1.13 A. In 50 of 80 cases, pLDDT exceeded 0.7 with RMSD under 1.5 A. Removing the cyclic offset significantly reduced accuracy.

Sequence redesign and hallucination

Across 3,274 unique backbone clusters, 63 Rosetta-designed sequences reached pLDDT above 0.9, compared with 1,145 from AfCycDesign. For hallucination, 48,000 models were generated per peptide size:

Seven X-ray structures (one redesigned, six hallucinated) matched their design models with RMSD below 1.0 A.

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 AfCycDesign on Tamarind Bio

• Macrocycle scaffold generation: hallucinate new 7 to 13 residue cyclic peptides as starting points for therapeutics.

• Sequence redesign: redesign sequences onto cyclic backbones from Rosetta or other samplers.

• Structure prediction: predict the fold of an existing cyclic peptide sequence and assess confidence with pLDDT and PAE.

• Candidate triage: filter large design sets by confidence before synthesis.

How to Use AfCycDesign on Tamarind Bio

1. Open the tool: log in to tamarind.bio and open AfCycDesign.

2. Choose a mode: prediction, redesign or hallucination, if exposed.

3. Provide inputs: a cyclic peptide sequence for prediction, a cyclic backbone for redesign, or a target length (the paper tests 7 to 13 residues) for hallucination.

4. Set parameters: adjust options such as the number of designs or an MSA, if exposed.

5. Run the job: submit and wait for completion.

6. Download outputs: retrieve predicted structures and confidence scores.

7. Interpret results: favor high pLDDT, but note that some accurate predictions had pLDDT below 0.85.

8. Go downstream: validate top candidates with energy calculations or experiments.

Things to Keep in Mind

• The prediction test set is small (80 NMR structures).

• Design and hallucination are limited to canonical L-amino acids.

• Disulfide connectivity is not explicitly constrained.

• Hallucinated designs were not tested for target binding or function.

• One hallucinated design (RH9.1) was not crystallized and its fold is unconfirmed.

Source: Rettie SA, Campbell KV, Bera AK, Kang A, Kozlov S, De La Cruz J, Adebomi V, Zhou G, DiMaio F, Ovchinnikov S, Bhardwaj G. Cyclic peptide structure prediction and design using AlphaFold. bioRxiv preprint, 2023. DOI: 10.1101/2023.02.25.529956.