Machine Learning-Assisted Discovery of Chiral Azobenzene Dopants for Liquid Crystals

Area of Science:

• Materials Science
• Organic Chemistry
• Computational Chemistry

Background:

• Cholesteric liquid crystals exhibit unique optical properties derived from their helical structures.
• Chiral dopants, specifically chiral azobenzene compounds, influence these helical structures via their helical twisting power (HTP).
• Photoisomerization of azobenzene derivatives allows for external control over HTP, enabling tunable optical properties.

Purpose of the Study:

• To develop and implement a machine learning (ML) workflow for the high-throughput screening of chiral azobenzene compounds.
• To identify novel azobenzene derivatives exhibiting a large change in helical twisting power (ΔHTP) for photoresponsive applications.
• To leverage molecular fingerprint feature importance analysis to guide the design of high-performance chiral dopants.

Main Methods:

• Training ML classification models using a dataset of 35 azobenzene compounds with known ΔHTP values.
• Performing virtual screening of 11,184 azobenzene derivatives to predict potential candidates.
• Utilizing molecular fingerprint feature importance analysis to understand structure-property relationships.

Main Results:

• Identification of a novel chiral azobenzene compound with a record high ΔHTP of 43 [1/μm].
• The synthesized compound demonstrated superior performance compared to existing materials in the training dataset.
• The ML workflow successfully predicted and guided the discovery of high-ΔHTP compounds.

Conclusions:

• The integration of machine learning and feature analysis provides a powerful and efficient strategy for designing functional chiral molecules.
• This approach accelerates the discovery of advanced materials with tailored photoresponsive properties.
• The developed workflow can be applied to the design of other functional organic materials.