Prediction And Understanding Of AIE Effect By Quantum Mechanics-aided Machine-learning Algorithm

Because of the fluorescence quenching in high concentration or in the aggregation state,the application of traditional organic luminescent materials is usually limited in many occasions.Fortunately,aggregation-induced emission(AIE)provides a revolutionary solution to the fluorescence self-quenching problem.For its unique emission characteristics of lower luminescence efficiency while dispersed in solutions and higher luminescence efficiency in the aggregate state,AIE materials show good application prospects in the basic life science,medicine,chemistry and other fields.However,the discovery of new AIE materials is driven mainly by laborious trial-and-error techniques.Therefore,it is of great significance to establish a model that can predict the AIE activity of organic luminescent materials prior to experiment.Currently,there are many different types of common AIE molecules,among which triphenylamine(TPA)derivatives have been widely studied in biomedical imaging,diagnosis and treatment,chemical sensing,organic light-emitting diodes and other fields in terms of applications due to their simultaneous AIE activity and TICT activity.However,many TPA derivatives are not AIE-active,and there is no effective method to accurately predict the AIE activity of TPA derivatives.Taking TPA derivatives as an example,an efficient mathematical model that can predict AIE activity was proposed by combining quantum mechanics and machine learning in this study.Considering that quantum mechanics can be used to calculate physical and chemical parameters of molecules,while machine learning possess ability to fit mathematical relations between variables from the multi-dimensional data of large sample size.Herein,61 kinds of TPA derivatives were collected from the literature,and their structures were optimized with DFT while the intramolecular charge distribution was calculated through NBO analysis.Then,support vector machine(SVM)classification algorithm was used to fit the relationship between charge distribution and AIE activity,so as to establish a machine learning model that can predict the AIE activity of TPA derivatives.The sensitivity,specificity and accuracy of the SVM classifier were optimized to reach 0.80,0.90 and 0.84,respectively,indicating a good prediction effect on AIE activity of TPA derivatives.On the analogy of dipole moment,this study also defined a physical quantity that can approximately describe the asymmetry of charge distribution on the TPA core,and found the optimal threshold value through the ROC curve,so as to establish a simple discriminant equation for AIE activity,whose classification effect for AIE activity is exactly the same as that of the SVM classifier.These results reveal that large enough dipole moment of TPA core is the key to activate the AIE effect of TPA derivatives,which agrees with the interpretation of the TICT mechanism.Finally,a novel AIE molecule was predicted by the established machine learning model,and the AIE activity was verified by experiments.This study provides new solutions for AIE research and high-throughput computational screening of AIE materials.