Prediction of Vertical Excitation and Emission Energies for Optoelectronic Molecules: An Automated Workflow Combining High-Throughput Computations and Machine Learning

Organic optoelectronic materials with localized or delocalized excitation features are widely used in various optoelectronic devices. A data-driven automated workflow was implemented for rapidly predicting excited state properties and screening of candidate molecules. A database was built with a collection of 1223 samples in a wide chemical space, including acceptor/donor complexes (such as PM6-Y6) and their derived subsystems. The ground state Dr index was demonstrated to be an important descriptor for evaluating the charge transfer tendency and predicting the Dr* index at the excited state through machine learning (ML) models. The ML-derived relationship among the Dr descriptor, excitation energy, and emission energy was applied to reproduce experimental results without the need for time-consuming calculations of excited states. The π-conjugated units containing three to six fused rings and N/S heteroatoms were found to have a stronger charge transfer tendency than the other functional units, highlighting the great potential of these chromophores in optoelectronic materials.