NCPepFold: Accurate Prediction of Noncanonical Cyclic Peptide Structures via Cyclization Optimization with...

Area of Science:

• Biomolecular science
• Computational chemistry
• Structural biology

Background:

• Current artificial intelligence (AI) methods for peptide structure prediction are limited to the 20 natural amino acids.
• Peptides with noncanonical amino acids are crucial for stability and function but are challenging to model.
• Existing models struggle with the unique structural properties of cyclic peptides.

Purpose of the Study:

• To develop a novel computational approach, NCPepFold, for predicting the structures of cyclic peptides containing noncanonical amino acids.
• To improve the accuracy and applicability of AI-driven peptide structure prediction.
• To facilitate the design of more stable and functional peptide-based therapeutics.

Main Methods:

• NCPepFold utilizes a specific cyclic position matrix for direct structure prediction.
• The model integrates multigranularity information at both residual and atomic levels.
• Fine-tuning techniques are employed to enhance prediction accuracy.

Main Results:

• NCPepFold demonstrates high accuracy in predicting cyclic peptide structures.
• The average peptide root-mean-square deviation (RMSD) achieved was 1.640 Å for cyclic peptides.
• The model successfully handles peptides with noncanonical amino acids.

Conclusions:

• NCPepFold is a novel deep learning model specifically designed for cyclic peptides with noncanonical amino acids.
• This approach significantly advances the field of peptide structure prediction.
• NCPepFold holds great potential for peptide drug design and biomedical research.