3.2 Virtual Library of Designed Polyesters
Traditional polyesters are mainly made by the condensation of diacids and diols. In silico retrosynthesis route enables the generation of a virtual library of 95,586 polyesters, i.e., dataset B, based on 695 original polyesters. From the computer-aided design point of view, we assumed that once the ester group is correctly present the corresponding polyester is yielded. To illustrate the relationship between datasets A and B more clearly, we then visualized the two datasets separately in two-dimensional chemical space, as shown in Figure 5a . It shows that the chemical diversity of the two datasets is quite similar, while compared to dataset A (blue points), dataset B (pink points) clearly covers more possible chemical structures of polyesters. This result implies these virtual polyesters significantly expand the chemical space of the existing polyesters and effectively overcomes the issue of data scarcity. In silico design of these virtual polyesters helps us to explore new polyesters with desired properties from a wider region of chemical space.
However, it is unrealistic to obtain over 95000 polyesters by experimental synthesis. By using the as-developed QSPR model,T gs of the hypothetical polyesters in dataset B were predicted. Figure 5b depicts the distributions ofT gs in datasets A and B. Similar trend in both datasets demonstrates that the hypothetical polyesters are almost consistent with the polyesters already presented in the database A. Therefore, a trustworthy virtual library for polyester has been successfully established, expanding the existing space for polyester materials and providing certain data support for the synthesis analysis of polyester materials.