A Set of Complementary Descriptors for the Power Conversion Efficiency Predictions of Organic Solar Cells

Organic
solar cells (OSCs) have undergone rapid development
over
the past few decades owing to their high efficiency, mechanical flexibility,
and potential for low-cost large-scale fabrication. Machine learning
(ML) models capable of accurately predicting the power conversion
efficiency (PCE) of OSCs can greatly accelerate the discovery of high-performance
donor and acceptor materials. In this work, we developed ML models
for PCE prediction by introducing a set of complementary molecular
descriptors and performing a comprehensive descriptor selection. The
resulting models achieved the highest test set R2 value of 0.82 and average R2 value
of 0.76 on an updated mixed experimental data set, representing one
of the best accuracies reported to date for OSC PCE prediction. Validation
using published experimental data confirmed the model’s strong
predictive accuracy and generalizability. Our results highlight that
the proposed OPT3D descriptors effectively complement the structural
information missing from RDKit and Morgan fingerprint descriptors.
This study demonstrates that the development of complementary molecular
descriptors is critical for improving ML model accuracy in molecular
material research and provides a promising approach for the rational
design of high-efficiency OSC materials.