High‐Throughput Deep Learning Prediction of D‐π‐A Dye Absorption Maxima on Titania Surfaces

Abstract The absorption maxima of D‐π‐Aorganic dyes on titania surfaces directly influence their ability toefficiently harvest sunlight in dye‐sensitized solar cells (DSSCs) andrelated technologies. Traditional methods of synthesizing and testing dyes aretime‐consuming.While theoretical TD‐DFT calculations are commonly used to estimateabsorption maxima, they require complex, computationally demanding clusterapproaches, making them impractical for high‐throughput screening. Accuratepredictions streamline the design process, saving both time and cost. In thiswork, we present a highly accurate and rapid deep‐learning approach thatleverages chemical fingerprint descriptors to predict the absorption maxima ofdyes on titania surfaces. Our method is trained on experimental data from over4000 organic chemical structures, ensuring robust and reliable performance. Ourensemble of 10 models achieved an R² score of 0.94, indicating an excellent, fast, and efficient predictive model capable of capturing complex relationshipsbetween molecular features and absorption behavior. To support the broaderscientific community, we have developed a freely available web‐based framework.This user‐friendly tool enables researchers to predict absorption maxima andperform high‐throughput screening of multiple and unknown chemical structuresdirectly from SMILES representations within seconds, serving as a significantasset for accelerating dye design.