Machine Learning-Based Screening for Potential Singlet Fission Chromophores: The Challenge of Imbalanced Data Sets

Area of Science:

• Materials Science
• Computational Chemistry
• Organic Electronics

Background:

• Singlet fission (SF) materials generate two triplet excitons from one photon, potentially doubling solar cell efficiency.
• SF molecules are promising for next-generation organic photovoltaics but are challenging to identify.
• Molecules with low-to-intermediate diradical character (DRC) are identified as potential SF chromophores.

Purpose of the Study:

• To develop a cost-effective strategy for discovering new SF chromophores using computational methods.
• To implement a machine learning-aided screening workflow based on molecular diradical character (DRC).
• To analyze structure-property relationships for designing novel SF materials.

Main Methods:

• Utilized a large dataset of 469,784 compounds from PubChem.
• Developed and applied machine learning classification models to predict SF potential based on DRC.
• Employed K-means clustering to analyze structure-property relationships of identified SF candidates.

Main Results:

• Successfully developed high-performing classification models for identifying potential SF chromophores.
• Identified a subset of compounds (approximately 4%) with high potential for singlet fission.
• Revealed qualitative structure-property relationships to guide future molecular design.

Conclusions:

• A machine learning-based screening approach effectively identifies potential singlet fission chromophores.
• The developed method and dataset facilitate the discovery of new SF materials for organic photovoltaics.
• The methodology is adaptable for diradicaloid applications in photonics and spintronics.