Efficient Adversarial Generation of Thermally Activated Delayed Fluorescence Molecules

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Deep learning models are crucial for molecular design and discovery.
• Exploring chemical space efficiently is key for developing new materials.
• Thermally activated delayed fluorescence (TADF) molecules are vital for efficient organic light-emitting diodes (OLEDs).

Purpose of the Study:

• To develop an integrated framework for designing novel TADF molecules.
• To leverage adversarial generative models for efficient exploration of chemical space.
• To identify TADF candidates with properties suitable for OLED applications.

Main Methods:

• Algorithmic synthesis of donor-acceptor (D-A) complexes using retrosynthetic rules.
• High-throughput labeling and prediction with deep neural networks.
• Adversarial autoencoder for generating new D-A molecules with target excited-state properties.
• Fine screening including spin-orbital coupling and excited-state optimization.

Main Results:

• Generated D-A molecules exhibit excited-state property distributions matching original samples.
• Created structures maintain significant charge transfer characteristics.
• Identified TADF candidates possess minimal adiabatic singlet-triplet gaps and moderate spin-orbital coupling.

Conclusions:

• The integrated framework effectively designs novel TADF molecules.
• The generated molecules show promise for use in efficient OLED devices.
• The approach facilitates the discovery of new materials within unexplored chemical spaces.