Unraveling the morphological complexity of two-dimensional macromolecules

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Two-dimensional (2D) macromolecules, including graphene and graphene oxide, exhibit diverse conformational phases.
• Current morphological classification relies on subjective visual inspection, limiting the understanding of underlying physics like deformation and surface contact.

Purpose of the Study:

• To develop an objective, data-driven method for classifying the morphologies of 2D macromolecules.
• To integrate physical principles into machine learning models for enhanced accuracy in phase characterization.

Main Methods:

• Utilized molecular simulations to generate 2D macromolecule samples.
• Extracted key features: metric changes, curvature, conformational anisotropy, and surface contact.
• Employed unsupervised learning for initial morphology classification (quasi-flat, folded, crumpled, interphases) and supervised learning for refinement.

Main Results:

• Successfully classified 2D macromolecule morphologies using geometrical and topological labels derived from simulations.
• Demonstrated that integrating physics-based features significantly improves the performance of machine learning models.
• Established a data-driven approach to characterize microstructures and molecular processes.

Conclusions:

• Machine learning offers a robust framework for objective classification of 2D macromolecule phases.
• The integration of physical insights enhances the predictive power of computational models.
• This approach provides a deeper understanding of the processing-microstructures-performance relationships in 2D materials.