Predicting Molecular Self-Assembly on Metal Surfaces Using Graph Neural Networks Based on Experimental Data Sets.

Area of Science:

• Supramolecular chemistry
• Materials science
• Surface science

Background:

• Molecular self-assembly on surfaces is crucial for nanotechnology.
• Traditional methods combining experiments with simulations are often laborious.
• Machine learning (ML) offers efficient and accurate prediction of molecular properties.

Purpose of the Study:

• To develop a machine learning model for predicting the self-assembly of functional polycyclic aromatic hydrocarbons (PAHs) on metal surfaces.
• To leverage experimental data for training and validating the ML model.
• To enable predictive design of nanostructures using functional molecules.

Main Methods:

• Construction of a graph neural network (GNN) model.
• Characterization of self-assembled nanostructures using scanning tunneling microscopy (STM).
• Training the GNN model on an experimental dataset of PAH molecules on metal surfaces.

Main Results:

• The developed GNN model demonstrated superior predictive performance compared to traditional ML algorithms.
• The model's generalization capability was verified through predictions on different functionalized molecules.
• Successful correlation between ML predictions and experimental results was achieved.

Conclusions:

• Experimental data can be used to train predictive ML models for molecular self-assembly.
• The developed ML model exhibits strong generalization performance.
• This approach facilitates the predictive design of functional nanostructures.