Improving an EVM QSPR model for glass transition temperature prediction using optimal design

An energy, volume and mass (EVM) model, involving four physico-chemical descriptor variables, i.e. van der Waals energy, internal energy, volume, and mass, to successfully predict the glass transition temperatures (Tg) of aliphatic acrylate and methacrylate polymers, has been previously described. The EVM model is as good, or better, as the previous models in terms of accuracy of calculated Tg values of polymers. However, the classical EVM approach is still limited by the validity of the experimental data that were used to derive the regressor coefficients of the quantitative structure–properties relationship (QSPR). In fact, one major problem is the large variation in the experimental Tg values that were reported in the literature. Deciding which values to use for modelling the relationship and the evaluation set of polymers is a problem to tackle with. For these reasons, an a priori design approach to the selection of a database of acrylate and methacrylate polymers for the evaluation of the EVM model has been adopted. In particular, the selection of the molecules to be considered was performed by two computed-assisted procedures based on the exchange algorithm for obtaining D-optimal design and the uniform method for finding experimental designs characterized by a stable structure. Based on the a priori design criteria, the selection of the optimal and uniform designs was “carried out”, in particular according to G-optimality.