Application of Machine Learning in Developing Quantitative Structure-Property Relationship for Electronic Properties...

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• The degree of π orbital overlap (DPO) model is a quantitative structure-property relationship (QSPR) tool for predicting electronic properties of polycyclic aromatic hydrocarbons (PAHs) and thienoacenes.
• Current DPO model limitations include inefficient manual parameter optimization and descriptor formulation, restricting its application scope.

Purpose of the Study:

• To develop a machine learning (ML)-based method for efficient DPO parameter optimization.
• To propose a simplified, truncated DPO descriptor for automatic extraction from molecular strings.
• To enhance the applicability of DPO models for high-throughput screening.

Main Methods:

• Implemented a machine learning (ML) approach to optimize DPO parameters.
• Developed a truncated DPO descriptor extractable from simplified molecular-input line-entry system (SMILES) strings.
• Validated the ML-optimized DPO model against experimental electronic properties (band gaps, electron affinities, ionization potentials).

Main Results:

• The ML-based methodology optimized DPO parameters using four times less data compared to previous methods.
• The ML-optimized DPO model achieved prediction accuracy within 0.1 eV for electronic properties.
• The truncated DPO model demonstrated comparable accuracy to the full DPO model.

Conclusions:

• The ML-based DPO approach significantly improves the efficiency of parameter optimization.
• The truncated DPO descriptor simplifies descriptor extraction and maintains predictive accuracy.
• This integrated approach facilitates automated pipelines for QSPR development and materials discovery.