Evolving Light Harvesting Metal Complexes with AI-Made Ligands

Light harvesting materials play a fundamental role in the development of photovoltaic technologies, including dye-sensitized solar cells. Transition metal complex (TMC) chromophores can push this field forward but their design is challenged by the need for optimizing multiple properties. An ideal chromophore would exhibit both intense and broad absorption in the visible range of the spectrum as well as, from a green chemistry perspective, high solubility in polar solvents including water. We hereby present a computational, data-driven approach to the discovery of novel TMC chromophores based on an evolutionary machine learning method combining elements of artificial intelligence (AI) and evolutionary computing (EC). In particular, AI-made bidentate ligands generated by a variational autoencoder were leveraged with an EC genetic algorithm (GA) for the multiobjective optimization of [RuL3]2+ chromophores. The fitness of the hits was consistent with intense, broad-spectrum absorption, and high solubility in polar solvents. The evolution of the absorption spectrum could be monitored step by step and easily interpreted by analyzing the frequency with which the ligands were selected by the GA. Based on the results, we suggest a set of experiments to the community doing wet lab research in this field.