Operando Evaluation of NIR-Absorbing Squaraine Molecules for Dye-Sensitized Solar Cells: An Informatics-Driven Molecular Machine Learning Framework

Near-infrared (NIR) absorbing dyes are crucial for enhancing the performance and transparency of dye-sensitized solar cells (DSSCs), offering a pathway toward more efficient, aesthetically compatible, and versatile photovoltaic systems. Identifying such dyes could represent a quantum leap in the field of photovoltaics, overcoming current limitations in efficiency and spectrum utilization. This study presents an informatics-driven molecular machine learning framework, leveraging operando methodologies to identify optimal NIR dyes. From a data set of 128 squaraine dyes, molecular descriptors were refined from 208 to 81 features using correlation and variance filtering. Among 42 models evaluated, a hybrid framework combining XGBoost and ExtraTrees Regressor achieved high predictive accuracy ( R 2 = 0.85, RMSE = 18.8, MAE = 12.24), capturing complex structure–property relationships. The framework identified NIR_D19, NIR_D18, and NIR_D15 as top-performing dyes. Molecular similarity analyses using Morgan fingerprints, Taylor-Butina clustering, and Tanimoto metrics revealed structural insights and key functional fragments influencing absorption. Validated against experimental data, this robust and scalable framework accelerates NIR dye discovery, advancing DSSC technologies for transparent and efficient photovoltaics.