Structure Prediction of Cyclic Peptides via Molecular Dynamics and Machine Learning

A major challenge in cyclic peptide development is the limited availability of structural information, which hinders structure-based design and makes it difficult to understand why different cyclic peptide sequences exhibit varying binding affinities, membrane permeabilities, and other properties. This lack of structural insight arises because most cyclic peptides exist as ensembles of multiple conformations in solution, making them extremely challenging to characterize experimentally using techniques such as solution NMR spectroscopy. In this talk, I will describe how molecular dynamics simulations with enhanced sampling methods can be used to obtain this critical structural information. I will also introduce StrEAMM (Structural Ensembles Achieved by Molecular dynamics and Machine learning), a framework that integrates simulation data with machine learning to rapidly generate high-quality structural predictions for cyclic peptides. This capability to efficiently predict structural ensembles will help researchers uncover the structural origins of cyclic peptide behavior and significantly accelerate the development of this promising class of molecules.