Deciphering Design of Aggregation-Induced Emission Materials by Data Interpretation.

This work presents a novel methodology for elucidating the characteristics of aggregation-induced emission (AIE) systems through the application of data science techniques. A new set of chemical fingerprints specifically tailored to the photophysics of AIE systems is developed. The fingerprints are readily interpretable and have demonstrated promising efficacy in addressing influences related to the photophysics of organic light-emitting materials, achieving high accuracy and precision in the regression of emission transition energy (mean absolute error (MAE) ∼ 0.13eV) and the classification of optical features and excited state dynamics mechanisms (F1score ∼ 0.94). Furthermore, a conditional variational autoencoder and integrated gradient analysis are employed to examine the trained neural network model, thereby gaining insights into the relationship between the structural features encapsulated in the fingerprints and the macroscopic photophysical properties. This methodology promotes a more profound and thorough comprehension of the characteristics of AIE and guides the development strategies for AIE systems. It offers a solid and overarching framework for the theoretical analysis involved in the design of AIE-generating compounds and elucidates the optical phenomena associated with these compounds.