Cyclic peptide structure prediction and design using AlphaFold.

Area of Science:

• Biomolecular structure prediction
• Computational chemistry
• Peptide drug design

Background:

• Cyclic peptides are promising therapeutics, but designing them computationally is challenging due to limited structural data.
• Existing deep learning models like AlphaFold excel at protein structure prediction but require adaptation for cyclic peptides.

Approach:

• Modified the AlphaFold deep learning network to predict cyclic peptide structures from amino acid sequences.
• Developed computational methods for designing new cyclic peptide sequences and backbones.
• Explored structural diversity of cyclic peptides ranging from 7 to 13 amino acids.

Key Points:

• The modified AlphaFold accurately predicted native cyclic peptide structures (36/49 cases with high confidence, RMSD < 1.5 Å).
• Generated approximately 10,000 unique cyclic peptide design candidates with high predicted folding confidence.
• Experimental validation (X-ray crystallography) confirmed atomic-level accuracy for seven designed cyclic peptides (RMSD < 1.0 Å).

Conclusions:

• The developed deep learning approach enhances cyclic peptide structure prediction accuracy.
• Novel computational methods facilitate the design of new macrocyclic peptides.
• This work provides a foundation for creating custom cyclic peptides for therapeutic applications.