Integrating Theory and Experiment with Graph Neural Networks to Classify Molecular Self-Assembly on Metal Surfaces

Area of Science:

• Materials Science
• Computational Chemistry
• Surface Science

Background:

• Predicting molecular self-assembly on surfaces is crucial for designing functional nanomaterials.
• Traditional methods often struggle with the complexity of molecule-surface interactions.

Purpose of the Study:

• To develop a predictive framework for molecular self-assembly on metal surfaces.
• To integrate computational (DFT) and experimental (STM) data for enhanced accuracy.

Main Methods:

• Constructed a dataset of 20 aromatic molecules on Au, Ag, and Cu substrates.
• Utilized density functional theory (DFT) derived descriptors.
• Employed a modified graph attention network (GAN) model trained on multisource data.

Main Results:

• Identified molecule-substrate interactions and charge transfer as key assembly factors.
• Achieved >95% accuracy in classifying molecular arrangements.
• Obtained an R-squared value of 0.985 for adsorption energy regression.
• Demonstrated model generalizability on unseen molecules.

Conclusions:

• Established a novel machine learning framework bridging computational and experimental data.
• The framework accurately predicts molecular self-assembly behavior.
• Paves the way for rational design of surface-supported nanostructures.