In this study, we present a new, generalized quantitative structure property relationship (QSPR) model for predicting the acentric factor utilizing a wide range of molecular species data from the DIPPR physical property database. Non-linear QSPR models involving robust artificial neural networks (ANN) and non-linear descriptor reduction techniques were developed.

The hypothesis for this work was to utilize an approach that calls for the use of cause-and-effect to determine to what extent a given descriptor accounts for the variations in molecular volume, area, shape, polarity, association (VASPA), etc. of a molecule, rather than attempting to model the properties using QSPR directly. The quality of the predictions obtained for this diverse group of molecules demonstrates the validity of an integrated approach and provides credible evidence to support the above hypothesis. Specifically, the ANN QSPR models were found to be capable of providing generalized a priori predictions for acentric factor with an absolute average deviation (AAD) of 8%. Further, the contributions of functional groups to acentric factor have also been identified.