Atom-Type-Based AI Topological Indices for Artificial Neural Network Modeling of Retention Indices of Monomethylalkanes

In this work, a combination of Xu and atom-type-based AI topological indices (TIs) were employed for quantitative structure-retention relationship (QSRR) study of monomethylalkanes (MMAs). A total of 196 temperature-programmed gas chromatographic retention indices corresponding to all C4-C30 MMAs on OV-1 stationary phase have been used in QSRR modeling. Results of the study showed that an artificial neural network (ANN) with 4-9-1 topology and Levenberg-Marquardt training algorithm can predict the retention indices with high degree of accuracy. The statistics of root-mean-square error for the training, validation and test sets were 0.200, 0.316 and 0.215, respectively. The proposed model resulted in a maximum relative error of 0.24% suggesting the TIs as excellent alternative for estimating retention indices of MMAs. According to the obtained results, relative importance of the TIs decreased in the order of AI(-CH3)> AI(-CH2-)> AI(>CH-)> Xu showing significant role of molecular branching, steric factor and molecular size as effective structural features on retention indices of MMAs.