Accurate structure prediction of cyclic peptides containing unnatural amino acids using HighFold3

HighFold3 is an enhanced model developed on the basis of AlphaFold 3, designed to accurately predict the three-dimensional structures of peptides containing unnatural amino acids (unAAs), including both monomers and their protein complexes. It is capable of handling special topological features such as head-to-tail cyclization and disulfide bond constraints. The overall architecture of HighFold3 is illustrated in Figure . HighFold3 introduces the Cyclic Position Offset Encoding Matrix (CycPOEM) and employs an innovatively designed “Cyclization Switch” module to dynamically select either a linear or cyclic positional encoding matrix within the model. When predicting cyclic peptide–protein complexes, the model explicitly divides the input distance matrix into two components: a linear positional encoding matrix for the target protein and a CycPOEM for the cyclic peptide ligand. This design enables the model to flexibly accommodate diverse conformational requirements and accurately model the binding of cyclic peptide ligands to protein receptors.

Model Architecture and Innovations:

HighFold3 is built upon the AlphaFold 3 framework, which is renowned for its accuracy in biomolecular structure prediction. The model introduces two key innovations:

Cyclic Position Offset Encoding Matrix (CycPOEM): This matrix adjusts positional encoding to account for the closed-loop structure of cyclic peptides, ensuring accurate representation of their topology. Cyclization Switch Module: This module dynamically selects between linear and cyclic positional encoding matrices, allowing flexibility in modeling diverse peptide conformations.

The architecture includes two sub-models: HighFold3-Linear for linear peptides and HighFold3-Cyclic for cyclic peptides, enabling tailored predictions based on peptide type.

Installation : Refer to the conda version of AlphaFold3 installation, and replace the feature files by highfold3/model.

Usage :

db_dir=af3_db_path

--model_dir=models_path

--json_path=json_path

--output_dir=output_path

--head_to_tail = 0 # or = 1

--disulfide_chain_res [[1,3,11]] # [[chain 1, Cys3, Cys11]]

Description :

1. The parameter head_to_tail indicates whether the head and tail form a ring, a boolean type.
2. The parameter disulfide_chain_res specifies the chain and positions where disulfide bonds are located.