Traditional topological indices vs electronic, geometrical, and combined molecular descriptors in QSAR/QSPR research

A comparison of the performance of molecular descriptors of different types was conducted. The study was concentrated on determining which descriptors are included in the best linear multivariate regression models constructed for modeling various physicochemical properties (melting point, boiling point, refractive index, molar volume, and density) and biological activities (anaesthetic activity, narcotic activity, sweetness intensity). A total of 84 molecular descriptors published over the past 2 decades was included in this study, plus some of their normalized and squared forms. It was shown that, for the estimation of physicochemical properties, the best small regression models with one to four parameters are mainly comprised of "classical" topological indices such as the Randić index, Wiener index, and molecular connectivity indices. For the correlation of biological activity, combinations of topological indices with geometrical descriptors produced regression models of the best quality. In-house developed software was used for generation of the molecular descriptors (the GROUND program) and for the statistical QSAR/QSPR analysis (the GROUNDSTAT program).