Exploring Molecular Descriptors and Acquisition Functions in Bayesian Optimization for Designing Molecules with Low...

Area of Science:

• Materials Science
• Organic Electronics
• Computational Chemistry

Background:

• Organic semiconductors offer potential for electronic devices but suffer from lower carrier mobility than inorganic counterparts like silicon.
• Hole reorganization energy is a critical factor influencing carrier mobility in organic semiconductors; lower energy correlates with higher mobility.
• Existing Bayesian optimization (BO) methods rely on various acquisition functions (AFs) whose performance is data-dependent.

Purpose of the Study:

• To identify organic semiconductor molecules with low hole reorganization energies using Bayesian optimization.
• To evaluate the performance of different acquisition functions (AFs) within BO for this specific application.
• To develop and propose a novel BO strategy that enhances search stability and efficiency.

Main Methods:

• Utilized Bayesian optimization (BO) as a machine learning framework for molecular discovery.
• Evaluated established acquisition functions (AFs) such as probability of improvement, expected improvement, and mutual information.
• Implemented and tested a novel approach involving alternating acquisition functions during the BO process.

Main Results:

• Demonstrated that the performance of acquisition functions (AFs) in Bayesian optimization (BO) is dataset-dependent for organic semiconductor molecules.
• Showcased that alternating acquisition functions (AFs) during the BO process leads to more stable and effective searches.
• Identified specific organic semiconductor molecules with potentially low hole reorganization energies.

Conclusions:

• Alternating acquisition functions (AFs) in Bayesian optimization (BO) provides a more robust and stable method for discovering organic semiconductor molecules with desirable electronic properties.
• This optimized search strategy can accelerate the development of high-performance organic electronic devices.
• Further exploration of adaptive acquisition function strategies is warranted for complex materials discovery problems.