Cyclic peptide structure prediction and design using AlphaFold

Abstract

Deep learning networks offer considerable opportunities for accurate structure prediction and design of biomolecules. While cyclic peptides have gained significant traction as a therapeutic modality, developing deep learning methods for designing such peptides has been slow, mostly due to the small number of available structures for molecules in this size range. Here, we report approaches to modify the AlphaFold network for accurate structure prediction and design of cyclic peptides. Our results show this approach can accurately predict the structures of native cyclic peptides from a single sequence, with 36 out of 49 cases predicted with high confidence (pLDDT > 0.85) matching the native structure with root mean squared deviation (RMSD) less than 1.5 Å. Further extending our approach, we describe computational methods for designing sequences of peptide backbones generated by other backbone sampling methods and for de novo design of new macrocyclic peptides. We extensively sampled the structural diversity of cyclic peptides between 7-13 amino acids, and identified around 10,000 unique design candidates predicted to fold into the designed structures with high confidence. X-ray crystal structures for seven sequences with diverse sizes and structures designed by our approach match very closely with the design models (root mean squared deviation < 1.0 Å), highlighting the atomic level accuracy in our approach. The computational methods and scaffolds developed here provide the basis for custom-designing peptides for targeted therapeutic applications.

Exploring a Structural Data Mining Approach to Design Linkers for Head-to-Tail Peptide Cyclization

“…Finally, we have also considered the very recent evolution of AlphaFold for cyclized peptides. 49 For 2ew4, 1ixt, 1m2c/1mii (GGAAGG linker), 2h8s, and 1m2c/ 1mii (GAGGAAG linker)), we obtain gRMSD 20 * values of 2.16, 1.49, 6.09, 3.32, and 3.50 Å, respectively, and lRMSD 20 * of 0.42, 0.50, 1.99, 2.70 and 1.91, respectively. The corresponding averages are 2.91 and 1.50.…”

“…Overall, these results are compatible with those presented for DaReUS-Loop, where we found Rosetta NGK performing better when starting from an experimental backbone, and less accurate when starting from a somewhat divergent backbone. Finally, we have also considered the very recent evolution of AlphaFold for cyclized peptides . For 2ew4, 1ixt, 1m2c/1mii (GGAAGG linker), 2h8s, and 1m2c/1mii (GAGGAAG linker)), we obtain gRMSD 20 * values of 2.16, 1.49, 6.09, 3.32, and 3.50 Å, respectively, and lRMSD 20 * of 0.42, 0.50, 1.99, 2.70 and 1.91, respectively.…”

“…In order to avoid this, it is necessary to understand the general sequence-structure requirements; in particular: what is the allowed sequence space of the linker and what will be the structure of the cyclized peptide? This is a challenging issue, and to the best of our knowledge, there is one computational protocol that has been successfully applied to head-to-tail cyclization linker design, namely the Rosetta protocol used by Bhardwaj et al In another recent study, the authors modified the AlphaFold framework for accurate design of cyclic peptides . However, in both studies, the sequence and structure of the entire cyclic peptide were designed from scratch.…”

“…This is a challenging issue, and to the best of our knowledge, there is one computational protocol that has been successfully applied to head-to-tail cyclization linker design, namely the Rosetta protocol used by Bhardwaj et al 6 In another recent study, the authors modified the AlphaFold framework for accurate design of cyclic peptides. 49 However, in both studies, the sequence and structure of the entire cyclic peptide were designed from scratch. At present, we do not know of any computational methods that can predict the sequence and structure of a head-to-tail cyclization linker while preserving the sequence and structure of the linear peptide that is being cyclized.…”

Training Neural Network Models Using Molecular Dynamics Simulation Results to Efficiently Predict Cyclic Hexapeptide Structural Ensembles

“…Recently, machine learning (ML) models such as AlphaFold2 and RoseTTAFold have shown exciting successes in predicting structures of folded proteins. − In theory, these models can be used to predict the structures of cyclic peptides, especially when the cyclic peptides have a clear fold and adopt structural motifs observed in folded proteins . However, it should be noted that cyclic peptides typically contain between 5 and 15 residues, , and thus they are generally unable to adopt regular secondary structures and most of the time lack a highly populated “fold”.…”