Cyclic peptide structure prediction and design using AlphaFold2.

Area of Science:

• Computational chemistry
• Biotechnology
• Drug discovery

Background:

• Small cyclic peptides are promising therapeutics.
• Developing deep learning models for cyclic peptide design is challenging due to limited training data.

Purpose of the Study:

• Introduce AfCycDesign, a deep learning approach for cyclic peptide design.
• Enable accurate structure prediction, sequence redesign, and de novo generation of cyclic peptides.

Main Methods:

• Developed AfCycDesign, a deep learning model for cyclic peptide design.
• Generated and validated de novo cyclic peptide sequences.
• Utilized designed peptides as scaffolds for developing binders.

Main Results:

• AfCycDesign identified over 10,000 structurally diverse cyclic peptide designs with high confidence.
• X-ray crystal structures of eight designed peptides closely matched computational models (RMSD < 1.0 Å).
• Designed peptide binders against MDM2 and Keap1 with nanomolar IC50 values.

Conclusions:

• AfCycDesign achieves atomic-level accuracy in cyclic peptide design.
• The developed computational methods and scaffolds facilitate custom peptide design for therapeutic applications.