Conclusions
In this contribution, we have successfully developed the QSPR model for evaluating the chemical structure of 695 polyesters with respect to their T g through a series of rigorous validations. Specifically, R 2 > 0.90 and Q 2LOO-CV = 0.88. Following this, a virtual library of nearly 100,000 polyesters has been built by in silico retrosynthesis, which greatly expands the available space for polyester materials. Their associatedT gs were predicted by the developed model as well. t-SNE shows significant chemical overlap with the known polyester database, i.e., dataset A, and the virtual library, i.e., dataset B, which demonstrates the rationality and feasibility of the design process.
Subsequently, 10 designed polyesters with differentT gs located in different temperature ranges were screened out for experimental synthesis. Good agreement between the experimental and predicted T gs not only demonstrates the accurate prediction performance of the QSPR model, but also verifies the efficiency of the design method. The rationality of the relationship between chemical structures andT gs was analyzed accordingly.
The methodology presented and the results gained in this work offer the potential to accelerate the design of high-performance polyesters, and may drive future product development of polymer industry.