Therapeutic cyclic peptides are attractive drug candidates because their ring-shaped structure can improve stability, binding strength, and membrane permeability. But one of the hardest steps in making them is deciding where to close the peptide chain, since the wrong cyclization site can sharply reduce yield.
Researchers have now introduced CycloPepper, a machine learning platform designed to predict cyclization outcomes for head-to-tail cyclic peptides. The system was trained on a standardized dataset generated with CycloBot, a fully automated synthesis platform that produced 306 cyclic peptides spanning 2 to 14 residues.
According to the study, the resulting model reached an average prediction accuracy of 84%. When the researchers tested the approach on 74 random and therapeutically relevant peptides, the predictions matched experimental results in 86% of cases.
The platform is meant to be practical for peptide chemists, offering both web-based and software access so users can rapidly assess possible cyclization sites before running synthesis. In validation work, it helped identify promising cyclization options for disease-targeting peptides, including sequences relevant to cancer biomarkers.
The broader goal is to reduce trial-and-error in cyclic peptide synthesis. By combining automation, standardized data generation, and machine learning, the team shows how computational tools may help streamline the development of peptide therapeutics.
For a field where synthesis success can depend heavily on sequence context, site selection, and conformational constraints, tools like CycloPepper could become an important part of the design workflow.
