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.