Quantitative structure–property relationships and neural networks: correlation and prediction of physical properties of pure components and mixtures from molecular structure

With the molecular structure of a molecule at hand the solution of the Schrödinger equation would allow the prediction of any physical, chemical or biological property in stationary states of molecules. However, although much progress has been made, particularly with semi-empirical methods, the practical application of quantum theory to complex molecules remains a distant possibility. As an alternative approach, QSPR (quantitative structure–property relationships) employ structural descriptors to develop correlations between the molecular structure and the physical property under investigation. High quality models have been developed for the normal boiling points of chlorosilanes, for the enthalpy of fusion of esters and for an equation of state mixture parameter for binary carbon dioxide–hydrocarbon systems using multilinear and nonlinear correlation techniques. The prediction capability for properties of compounds not present in the training set proved to be excellent for all properties correlated in this study, mostly within the accuracy of experimental measurements.