Integrating Ultra-Coarse-Grained Protein Models into Accessible Workflows for Multiscale Molecular Dynamics.

Area of Science:

• Computational Biology
• Biophysics
• Molecular Dynamics

Background:

• Molecular dynamics (MD) simulations require multiple resolutions for protein conformational analysis.
• All-atom (AA) simulations offer high resolution but are computationally expensive for large systems and long durations.
• Coarse-grained (CG) and ultra-coarse-grained (UCG) models reduce computational cost while preserving key protein features.

Purpose of the Study:

• To develop a less computationally intensive method for exploring protein conformational space.
• To integrate UCG models into the Multiscale Machine-learned Modeling Infrastructure (MuMMI) workflow.
• To enable accurate sampling of protein conformations, specifically for RAS-RAF protein interactions.

Main Methods:

• Integration of UCG models based on heterogeneous elastic network modeling (hENM) into MuMMI.
• Refinement of UCG intramolecular interactions using higher-resolution CG Martini simulation data.
• Development of machine learning-based backmapping methods using diffusion models to map between UCG and CG Martini structures.
• Creation of a Python package to estimate UCG bond coefficients from CG Martini simulation fluctuations.

Main Results:

• UCG models accurately sample protein conformations, demonstrated in RAS-RAF simulations.
• A scalable Python package was developed for refining UCG models.
• Novel machine learning backmapping techniques were implemented to recover detailed structures.
• UCG-mini-MuMMI, a computationally efficient version of MuMMI, was created.

Conclusions:

• UCG models offer an effective strategy for reducing computational costs in MD simulations.
• The developed UCG-mini-MuMMI provides an accessible resource for the scientific community.
• This approach is broadly applicable to various protein systems, offering insights into UCG model advantages and limitations.