Machine Learning Prediction of Optical Properties of Coumarin Derivatives Using Gaussian-Weighted Graph Convolution...

Area of Science:

• Computational Chemistry
• Materials Science
• Photochemistry

Background:

• Coumarin derivatives are vital as chromophores and fluorophores.
• Predicting their optical properties is crucial for various applications.
• Existing methods may not fully capture complex molecular interactions.

Purpose of the Study:

• To develop a machine learning model for predicting optical properties of coumarin derivatives.
• To introduce novel molecular representations: Gaussian-weighted graph convolution (GWGC) and subgraph modular input (SMI).
• To understand how substituents influence the optical properties of the coumarin core.

Main Methods:

• Constructed an experimental database of coumarin derivative optical properties (absorption and emission wavelengths).
• Developed machine learning models utilizing GWGC for interatomic effects and SMI for modular representation (core and substituents).
• Compared GWGC and SMI models against RDKit descriptors and Morgan fingerprint.

Main Results:

• The developed ML models based on GWGC and SMI demonstrated superior performance in predicting optical properties.
• GWGC effectively accounts for interatomic effects, enhancing prediction accuracy.
• SMI successfully modularizes molecular information, clarifying substituent influences.

Conclusions:

• Machine learning models employing GWGC and SMI offer a powerful and accurate approach for predicting coumarin derivative optical properties.
• This methodology provides insights into structure-property relationships.
• The GWGC and SMI approach is broadly applicable to molecules with core structures and substituents.