Use HighFold Online
Commercially Available HighFold No-Code Web Server
HighFold: AlphaFold-Based Structure Prediction for Cyclic Peptides and Peptide-Protein Complexes
HighFold is an AlphaFold-based method for predicting the 3D structures of cyclic peptides and their complexes, described in a 2023 bioRxiv preprint. It has two modules: HighFold_Monomer, built on AlphaFold, for single cyclic peptides, and HighFold_Multimer, built on AlphaFold-Multimer, for cyclic peptide-protein complexes. Cyclic peptides can be closed head to tail, stabilized by disulfide bridges, or both.
Cyclic peptides are attractive as drug candidates, but their ring constraints make structure prediction harder than for linear sequences. Standard AlphaFold-style models assume a linear chain, and docking tools can struggle to place a flexible cyclic ligand. HighFold addresses this by teaching the AlphaFold framework about the cyclic connectivity of the peptide.
Fact | Detail |
|---|---|
Task | Predict structures of cyclic peptides, with or without disulfide bridges, and cyclic peptide-protein complexes |
Modules | HighFold_Monomer (AlphaFold) and HighFold_Multimer (AlphaFold-Multimer) |
Key idea | A cyclic connectivity matrix that encodes head-to-tail and disulfide links |
Model ranking | Top-5 models ranked by pLDDT (monomer) or a pTM/ipTM combination (multimer) |
Benchmarks | 63 monomers (12 to 39 residues) and 17 ADCP complexes, plus 17 external PDB complexes |
How HighFold Works: Cyclic Connectivity Matrices and Disulfide Enumeration
Cyclic connectivity matrix: an N by N matrix, where N is the peptide length, encodes the head-to-tail ring and disulfide connections. It is built with the FCP algorithm, adapted from Floyd's shortest-path algorithm.
Disulfide enumeration: when two or more cysteines are present, the CSDB algorithm lists the possible cysteine pairings and builds a matrix for each combination.
Multimer matrix: for complexes, a block-diagonal matrix combines the matrices of each chain.
Ranking: the top-5 models are ranked by pLDDT for monomers or by a combined pTM and ipTM confidence for complexes.
HighFold Results: Cyclic Peptide Monomers and Peptide-Protein Complexes
Monomers Compared with AfCycDesign
Accuracy: median RMSD of 1.058 Å and average of 1.478 Å, with 37 of 63 cases under 1.5 Å.
Improvement: an average gain of 0.259 Å over AfCycDesign.
Two-disulfide peptides: 1.720 Å for HighFold versus 3.256 Å for AfCycDesign.
Complexes Compared with ADCP
Method | Average F_nat |
|---|---|
HighFold_Multimer | 0.813 |
ADCP | 0.415 |
Ligand RMSD for HighFold_Multimer ranged from 0.218 to 1.044 Å, with an average of 0.359 Å. On the external set of 17 PDB complexes, two examples (6u22 and 6bvh) reached an F_nat of 0.948.
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 HighFold on Tamarind Bio
Cyclic peptide structure prediction: model head-to-tail cyclic peptides and peptides with disulfide bridges.
Peptide-protein complex modeling: predict how a cyclic peptide binds a target protein.
Disulfide topology screening: compare models for different cysteine pairings.
Candidate triage: use confidence scores to rank cyclic peptide designs.
How to Use HighFold on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and open HighFold.
Choose the mode: select monomer or multimer prediction, if exposed.
Enter sequences: provide the cyclic peptide sequence and, for complexes, the target protein sequence.
Specify cyclization: indicate head-to-tail cyclization and disulfide options, if exposed.
Run the job: submit the run and wait for the models.
Download outputs: retrieve the top-ranked structures and confidence scores.
Interpret results: use pLDDT for monomers and pTM/ipTM-based confidence for complexes, remembering that confidence and RMSD are not strictly linear.
Use downstream: take the models into analysis, redesign or further simulation.
Things to Keep in Mind
There is no direct linear relationship between pLDDT and RMSD, although higher pLDDT tends to go with lower RMSD.
Accuracy drops for peptides longer than about 30 residues, which contain more random coil.
The complex benchmark has limited diversity, with integrase well represented, and few ligands of 10 or more residues with secondary structure.
Some cases were not satisfactory, for example one complex with an F_nat of 0.584.
Source: HighFold preprint, bioRxiv, 2023, DOI 10.1101/2023.08.27.554979 (https://www.biorxiv.org/content/10.1101/2023.08.27.554979v1). Author list and code link were not available from the fetched page.