Fast Generation of Simulation-Quality Structural Ensembles of Mixed-Chirality Cyclic Peptides via Diffusion Models.

Area of Science:

• Computational chemistry and drug discovery.
• Structural biology and biophysics.

Background:

• Cyclic peptides are a promising therapeutic class, but their design is complex.
• Many cyclic peptides exist as multiple conformations (structural ensembles) in solution, which is crucial for their function, including membrane permeability and molecular binding.
• Predicting these conformational ensembles is vital for advancing de novo cyclic peptide drug design.

Purpose of the Study:

• To introduce diffusion models for efficient and accurate prediction of cyclic peptide structural ensembles.
• To enable better computational design of cyclic peptide therapeutics by understanding their conformational flexibility.

Main Methods:

• Diffusion models were trained directly on molecular dynamics (MD) simulation data.
• Cyclic peptide structures were represented using sine and cosine values of backbone dihedral angles.
• Each frame from MD simulations served as a training instance.

Main Results:

• The diffusion models can generate high-quality cyclic peptide structures comparable to MD simulations.
• Generated structures accurately reflect the Boltzmann distribution sampled in MD simulations.
• This enables a deeper understanding of the physicochemical properties influencing cyclic peptide behavior.

Conclusions:

• Diffusion models offer an efficient and accurate method for predicting cyclic peptide structural ensembles.
• This approach facilitates the computational design of novel cyclic peptide therapeutics.
• Understanding conformational flexibility is key to harnessing the full potential of cyclic peptides in medicine.