High‐Throughput Structural Screening and Predictive Modeling of Organic Solar Cells via Machine Learning

The study develops and validates an integrated machine learning framework to accelerate the discovery of high‐performance Y6‐series nonfullerene acceptors for organic solar cells. Addressing the lack of models that directly link molecular structure to device‐level efficiency, especially for Y6‐based systems, the work investigates whether an end‐to‐end, interpretable pipeline can reliably predict power conversion efficiency (PCE) from molecular information and enable high‐throughput virtual screening. The resulting model achieves a Pearson correlation coefficient of 0.873 and R 2 = 0.762 on the test set, and maintains robust performance on an external dataset of 28 unseen pairs with r = 0.848 and R 2 = 0.682, reducing prediction time from days with conventional density functional theory (DFT) to minutes. Among virtual library of 29 620 Y6‐derived acceptors and 88 860 donor–acceptor combinations, the framework identifies 98 candidate pairs with predicted PCE above 20%, including a top candidate reaching 21.49%. Analysis confirms that the leading acceptor exhibits favorable frontier orbital alignment, reduced bandgap, and lower reorganization energy relative to benchmark Y6, suggesting enhanced charge transport and photocurrent generation. This work establishes a generalizable, interpretable, and computationally efficient approach for structure–property–performance prediction, providing a scalable pathway for data‐driven discovery of next‐generation organic photovoltaic materials and other organic electronic systems.