QSPR modeling: Graph connectivity indices versus line graph connectivity indices

Five QSPR models of alkanes were reinvestigated. Properties considered were molecular surface-dependent properties (boiling points and gas chromatogra phic retention indices) and molecular volume-dependent properties (molar vo lumes and molar refractions). The vertex- and edge-connectivity indices wer e used as structural parameters. In each studied case we computed connectiv ity indices of alkane trees and alkane line graphs and searched for the opt imum exponent. Models based on indices with an optimum exponent and on the standard value of the exponent were compared. Thus, for each property we ge nerated six QSPR models (four for alkane trees and two for the correspondin g line graphs). In all studied cases QSPR models based on connectivity indi ces with optimum exponents have better statistical characteristics than the models based on connectivity indices with the standard value of the expone nt. The comparison between models based on vertex- and edge-connectivity in dices gave in two cases (molar volumes and molar refractions) better models based on edge-connectivity indices and in three cases (boiling points for octanes and nonanes and gas chromatographic retention indices) better model s based on vertex-connectivity indices. Thus, it appears that the edge-conn ectivity index is more appropriate to be used in the structure-molecular vo lume properties modeling and the vertex-connectivity index in the structure -molecular surface properties modeling. The use of line graphs did not impr ove the predictive power of the connectivity indices. Only in one case (boi ling points of nonanes) a better model was obtained with the use of line gr aphs.