Prediction of Glass Transition Temperature (T_g) of Some Compounds in Organic Electroluminescent Devices with Their Molecular Properties

We have studied the quantitative structure−property relationship between descriptors representing the molecular structure and glass transition temperature (T_g) for 103 molecules including organic electroluminescent (EL) devices materials. Eighty-six descriptors were introduced and among them seven descriptors (one topological descriptor, one thermodynamic descriptor, one spatial descriptor, one structural descriptor, and three electrostatic descriptors) were selected by Genetic Algorithm (GA). The 81 molecules chosen randomly among 103 compounds were used as a training set, and the remaining 22 molecules were used as a prediction set. The quantitative relationship between these seven descriptors and T_g was tested by multiple linear regression (MLR) and artificial neural network (ANN). ANN analysis showed no significant advantage over MLR for this study. As the results of the MLR, the square of the correlation coefficient (R²) for the T_g of the 81 training set was 0.989, and the average error was 8.8 K. In prediction for T_g using the 22 prediction compounds set with MLR, R² was 0.976, and the average error was 13.9 K.