StrEAMM-Thioether: Efficient Structure Prediction for Thioether-Linked Cyclic Peptides.

Area of Science:

• Computational chemistry
• Biophysics
• Medicinal chemistry

Background:

• Cyclic peptides are promising therapeutics targeting protein-protein interactions and cell membranes.
• Predicting cyclic peptide structural ensembles is crucial for rational drug design but challenging experimentally.
• Existing computational methods like molecular dynamics (MD) are computationally intensive.

Purpose of the Study:

• To expand the StrEAMM (Structural Ensembles Achieved by Molecular Dynamics and Machine Learning) platform for thioether-linked cyclic peptides.
• To develop a fast and accurate computational method for predicting structural ensembles of thioether-linked cyclic peptides.
• To facilitate the design and discovery of cyclic peptide therapeutics compatible with mRNA display screening.

Main Methods:

• Developed and trained graph neural network models within the StrEAMM framework.
• Applied the models to predict structural ensembles of thioether-linked cyclic peptides.
• Experimentally synthesized and characterized predicted cyclic peptides using solution Nuclear Magnetic Resonance (NMR).

Main Results:

• The StrEAMM-thioether models accurately predict structural ensembles for thioether-linked cyclic peptides.
• Four thioether-linked cyclic pentapeptides were identified as well-structured based on model predictions.
• Experimental NMR characterization showed general agreement with the computationally predicted structures.

Conclusions:

• The StrEAMM-thioether models provide a fast and reliable method for predicting cyclic peptide structural ensembles.
• This approach streamlines the design and discovery of cyclic peptide therapeutics.
• The models can be integrated with mRNA display platforms for efficient drug discovery.