Prediction of Gas-Chromatographic Retention Indices Using Topological Descriptors
Matevž Pompe, Marjana Novič · Journal of Chemical Information and Computer Sciences · 1998
Theoretical prediction of gas-chromatographic retention indices could be used as an additional method for the identification of organic substances during gas-chromatographic separation. Our previously developed model, based on artificial neural networks, has been extended with the additional topological structural descriptors to improve prediction capabilities. The topological indices were selected for the representation of chemical structures because of their simplicity; therefore they could also be used for solving identification problems by chromatographers who are not experts in structural representation. An extensive data set of 381 simple organic compounds with known retention indices taken from the literature served as a training and test set. Sixteen informational and topological structural descriptors were selected for the description of molecular structure. The same data set was used for the prediction of gas-chromatographic retention indices using a multiple linear regression model and back-propagation of error and counterpropagation artificial neural network. The average root mean squared error values of a 10-fold cross-validation procedure were 22.5, 19.2, and 36.1, respectively.