Atom-Based Machine Learning Model for Quantitative Property-Structure Relationship of Electronic Properties of...

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Predicting electronic properties of organic molecules is crucial for materials design.
• Quantitative structure-property relationship (QSPR) models offer a data-driven approach.
• Fusenes and their derivatives are important organic semiconductors.

Purpose of the Study:

• To develop accurate machine learning-based QSPR models for predicting electron affinity, ionization potential, and band gap.
• To evaluate the performance of atom-based Weisfeiler-Lehman (WL) graph kernels combined with Gaussian process regression (GPR).
• To assess the utility of active learning for diverse chemical datasets.

Main Methods:

• Employed three variants of the atom-based Weisfeiler-Lehman (WL) graph kernel method.
• Utilized the Gaussian process regressor (GPR) machine learning model.
• Computed molecular properties using density functional theory (DFT) at the B3LYP-D3/6-31+G(d) level.

Main Results:

• Achieved accurate predictions for electronic properties of polycyclic aromatic hydrocarbons (PAHs) and their derivatives.
• Demonstrated low root-mean-square deviations (0.15 eV) for predicted electronic properties.
• Showcased the effectiveness of active learning for diverse and complex datasets.

Conclusions:

• GPR/WL kernel methods provide a robust framework for predicting electronic properties of fusenes.
• Model interpretability offers insights into the underlying chemical factors governing electronic properties.
• The developed QSPR models are valuable tools for accelerating the discovery of new organic electronic materials.