HEroBM: A deep equivariant graph neural network for high-fidelity backmapping from coarse-grained to all-atom...

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Biophysics

Background:

• Molecular simulations are crucial for studying dynamic properties in chemistry, biology, and materials science.
• Coarse-grained (CG) techniques simplify complex systems for large-scale simulations but lose atomistic detail.
• Current backmapping methods have limitations in accuracy, transferability, and reliance on energy relaxation.

Purpose of the Study:

• To develop a dynamic, scalable, and accurate method for reconstructing atomistic coordinates from coarse-grained simulations.
• To overcome the limitations of existing rule-based and machine learning backmapping approaches.
• To provide a versatile tool applicable to diverse chemical systems and scales.

Main Methods:

• Utilized deep equivariant graph neural networks in a hierarchical approach.
• Developed the HEroBM (Hierarchical Equivariant Reconstruction backMapper) method.
• Demonstrated applicability across various CG mappings and system compositions.

Main Results:

• Achieved high-resolution backmapping with superior accuracy compared to existing methods.
• Successfully reconstructed atomistic structures for complex biological systems, including a GPCR in a lipid bilayer.
• Showcased the method's ability to handle diverse CG mappings and system sizes.

Conclusions:

• HEroBM offers a versatile and efficient protocol for high-fidelity molecular structure reconstruction.
• The method enables seamless transitions between coarse-grained and all-atom simulations.
• This facilitates deeper insights into molecular mechanisms across various scientific domains.