Predicting Emission Wavelengths and Quantum Yields of Diverse Bis-cyclometalated Iridium(III) Complexes Using Machine Learning

Cyclometalated iridium(III) complexes are excellent emitters for phosphorescent organic light-emitting diodes, but the design of such complexes require substantial cost and experimental efforts. In turn, the predictive power of density functional theory calculations is seldom enough for reliable prediction of the excited state properties of iridium(III) complexes. In this work, we aimed at data-driven prediction of the emission energies and photoluminescence quantum yields of such complexes. To this end, we created a database (IrLumDB) that contains experimentally measured luminescence properties for over 1200 literature bis-cyclometalated iridium(III) complexes. Based on this database, we developed machine learning models that are capable of predicting the energy of emission maxima and photoluminescence quantum yields for the iridium phosphors with mean absolute errors of 18.26 nm and 0.129, respectively, requiring only SMILES of ligands. Furthermore, we validated the model for emission wavelength prediction on the set of 33 experimentally obtained luminescence spectra for newly synthesized and characterized iridium(III) complexes. Our data-driven methodology will complement quantum chemical calculations as an efficient alternative approach for the prediction of the excited-state properties of large sets of bis-cyclometalated iridium(III) complexes, facilitating computational discovery of efficient emitters. The emission properties prediction and the dataset exploration are available at https://irlumdb.streamlit.app/.