Molecular generation targeting desired electronic properties via deep generative models.

Area of Science:

• Materials Science
• Computational Chemistry
• Organic Electronics

Background:

• Discovering new functional materials requires exploring vast chemical spaces of precursor building blocks.
• Traditional methods may not identify non-obvious building blocks beyond chemical intuition.
• Artificial intelligence (AI) offers a powerful approach to generate diverse organic building blocks.

Purpose of the Study:

• To demonstrate the application of deep recurrent neural networks (DRNNs) for exploring chemical space.
• To generate novel donor-acceptor oligomers with specific electronic properties for functional materials.
• To identify non-obvious building blocks beyond conventional chemical expertise.

Main Methods:

• Utilized deep recurrent neural networks (DRNNs) for generative modeling.
• Trained the DRNN on a dataset of donor-acceptor oligomers.
• Employed atomic substitutions (e.g., halogenation, methylation) and molecular features (e.g., size) for tuning properties.
• Sampled from different subsets of the training database to enrich for desired properties.

Main Results:

• Successfully generated approximately 1700 new donor-acceptor oligomers.
• The DRNN learned to balance atomic substitutions and molecular features to produce novel structures.
• Generated oligomers were tuned for specific electronic properties, including a HOMO-LUMO gap <2 eV and a dipole moment <2 Debye.
• Demonstrated enrichment of the donor-acceptor library towards targeted properties.

Conclusions:

• AI, particularly DRNNs, is effective for exploring chemical space and generating novel organic building blocks.
• This method enables the discovery of non-obvious molecular designs for functional materials.
• The generated donor-acceptor oligomers show potential for applications in organic photovoltaics.