Enhanced Coarse-Grained Molecular Dynamics Simulation with a Smoothed Hybrid Potential Using a Neural Network Model

Area of Science:

• Computational Biology
• Biophysics
• Artificial Intelligence in Science

Background:

• All-atom (AA) molecular dynamics (MD) simulations face challenges in reproducing biomolecular structural changes due to rugged energy profiles and long timescales.
• Existing coarse-grained (CG) models often oversimplify energy landscapes, limiting exploration of metastable states distant from the initial structure without bias.

Purpose of the Study:

• To develop a hybrid potential combining artificial intelligence (AI) and coarse-grained (CG) methods to accelerate biomolecular dynamics simulations.
• To enable exploration of transitions between metastable states while preserving essential protein characteristics.

Main Methods:

• Developed a hybrid potential integrating an AI potential with a minimal CG potential (statistical bond length, excluded volume).
• Trained the AI potential using energy matching against AA force field energies from diverse structures sampled via multicanonical (Mc) MD simulations.
• Smoothed the energy profile via energy minimization before applying it to CGMD simulations.

Main Results:

• The AI potential demonstrated high accuracy in predicting AA energies (R-value > 0.89) for chignolin and TrpCage.
• CGMD simulations using the smoothed hybrid potential significantly enhanced transition dynamics between metastable states.
• The enhanced dynamics preserved protein properties compared to conventional CGMD and AAMD methods.

Conclusions:

• The developed AI-CG hybrid potential effectively accelerates exploration of biomolecular structural dynamics.
• This approach overcomes limitations of traditional MD and CG methods for studying transitions between distant metastable states.
• The methodology offers a promising tool for advancing computational studies in structural biology and drug discovery.