Backmapping coarse-grained macromolecules: An efficient and versatile machine learning approach

Area of Science:

• Polymer Science
• Computational Chemistry
• Materials Science

Background:

• Multiscale modeling of polymers requires linking coarse and fine molecular representations.
• Generating fine-scale details from coarse representations is challenging, demanding methods balancing accuracy and efficiency.
• Existing techniques often struggle with general applicability and computational cost.

Purpose of the Study:

• To develop an image-based approach for structural backmapping from coarse-grained to atomistic polymer models.
• To utilize machine learning, specifically conditional generative adversarial networks, for this backmapping process.
• To demonstrate the approach's efficiency and transferability using cis-1,4 polyisoprene melts.

Main Methods:

• Training conditional generative adversarial networks on paired coarse-grained and atomistic configurations.
• Applying the trained model to predict atomistic structures from coarse-grained inputs.
• Investigating the impact of different data representation schemes on model performance.

Main Results:

• Achieved efficient and accurate structural backmapping from coarse-grained to atomistic models for cis-1,4 polyisoprene melts.
• Demonstrated remarkable transferability across different molecular weights using training sets from oligomeric compounds.
• The image-based machine learning approach proved effective in generating high-fidelity atomistic configurations.

Conclusions:

• The proposed image-based backmapping approach offers a versatile and efficient solution for multiscale polymer modeling.
• This method facilitates the generation of high-fidelity initial configurations with minimal human intervention.
• The approach is readily extendable to other complex polymer systems.