Machine Learning Models for Predicting Optoelectronic Properties

Researchers at Middle East Technical University in Ankara, Turkey have developed machine learning (ML) models to predict key optoelectronic properties of conducting polymers. According to the study, the performance and reliability of ML-quantitative structure-property relationship (QSPR) models depend on the quality, size, and diversity of the data set used for model training. The team manually curated a large-scale data set containing 3120 donor-acceptor conjugated polymers, selecting the most utilized 60 donors and 52 acceptors. This data set serves as a valuable resource for ML-based prediction of key electronic properties such as band gap energy and hole reorganization energy. The study found that kernel partial least-squares (KPLS) regression utilizing radial and molprint2D fingerprints achieved the highest accuracy in predicting band gap energy, with values of 0.899 and 0.897, respectively. The developed ML models provide a predictive framework for high-performance organic photovoltaics (OPV) materials design, significantly reducing the reliance on labor-intensive experimental procedures and computationally expensive first-principle calculations.

Key Takeaways:

• The quality, size, and diversity of the data set used for model training significantly impacts the performance and reliability of ML-QSPR models.
• A large-scale data set containing 3120 donor-acceptor conjugated polymers was manually curated by selecting the most utilized 60 donors and 52 acceptors.
• Kernel partial least-squares (KPLS) regression utilizing radial and molprint2D fingerprints achieved the highest accuracy in predicting band gap energy, with values of 0.899 and 0.897, respectively.
• The developed ML models provide a predictive framework for high-performance OPV materials design, significantly reducing the reliance on labor-intensive experimental procedures and computationally expensive first-principle calculations.
• The study identified the importance of feature selection and data set optimization for accurate target property prediction in organic electronics.
• The research highlights the potential of ML to advance OPV research and development.

Statistics:

• 3120: number of donor-acceptor conjugated polymers in the curated data set
• 60: number of most utilized donors selected for the data set
• 52: number of most utilized acceptors selected for the data set
• 0.899: accuracy of KPLS regression using radial fingerprint for predicting band gap energy
• 0.897: accuracy of KPLS regression using molprint2D fingerprint for predicting band gap energy
• 0.830: accuracy of ML models for predicting hole reorganization energy

Sources:

• Band Gap and Reorganization Energy Prediction of Conducting Polymers by the Integration of Machine Learning and Density Functional Theory. Journal of Chemical Information and Modeling, 2025.
• Amer Chemical Soc, 1155 16TH St, NW, Washington, DC 20036, USA. (American Chemical Society - www.acs.org; Journal of Chemical Information and Modeling - www.pubs.acs.org/journal/jcisd8)
• Erol Yildirim, Dept. of Chemistry, Middle East Technical University, 06800 Ankara, Turkey.
• NewsRx. Middle East Technical University Reports Findings in Machine Learning (Band Gap and Reorganization Energy Prediction of Conducting Polymers by the Integration of Machine Learning and Density Functional Theory). Information Technology Newsweekly. June 10, 2025; p 430.