Enhanced Prediction of Absorption and Emission Wavelengths of Organic Compounds through Hybrid Graph Neural Network Architectures

The prediction of absorption and emission spectra is a crucial task in the fields of chemistry and materials science, traditionally approached through computational methods such as density functional theory (DFT). Although DFT provides accurate results, its high computational cost and extended processing time limit scalability for large molecular systems. In this study, we demonstrate the effectiveness of graph neural networks (GNNs) as a powerful alternative for predicting optical properties. Our approach leverages multiple GNN architectures that integrate both molecular graphs and molecular fingerprints, enabling the model to capture intricate structural and electronic features of molecules. The combination of molecular representations allows our GNN frameworks to efficiently discern relevant patterns and relationships, significantly improving prediction accuracy while requiring substantially less computational resources. This research underscores the potential of GNNs to revolutionize the approach to spectral predictions, offering a promising pathway for future investigations in molecular design and materials discovery