Cyclization provides a general strategy for improving the proteolytic stability, cell membrane permeability and target binding affinity of peptides. Insertion of a stable, non-reducible linker into a disulphide bond is a commonly used approach for cyclizing phage-displayed peptides. However, among the vast collection of cysteine reactive linkers available, few provide the selectivity required to target specific cysteine residues within the peptide in the phage display system, whilst sparing those on the phage capsid. Here, we report the development of a cyclopropenone-based proximity-driven chemical linker that can efficiently cyclize synthetic peptides and peptides fused to a phage-coat protein, and cyclize phage-displayed peptides in a site-specific manner, with no disruption to phage infectivity. Our cyclization strategy enables the construction of stable, highly diverse phage display libraries. These libraries can be used for the selection of high-affinity cyclic peptide binders, as exemplified through model selections on streptavidin and the therapeutic target αvβ3.
Representation learning is making inroads into drug discovery. A study in Nature Communications emphasizes multiple limitations in property prediction. The results suggest that continued research and improvements are required for this specific area that coalesces machine learning and molecular medicine. Machine learning is a powerful tool for the study and design of molecules. Here the authors comment a recent publication in Nature Communications which highlights the challenges of different molecular representations for data-driven property predictions.