Comparative study of different structural descriptors and variable selection approaches using partial least squares in quantitative structure-activity relationships

Marengo, E., Carpignano, R., Savarino, P. and Viscardi, G., 1992. Comparative study of different structural descriptors and variable selection approaches using partial least squares in quantitative structure-activity relationships. Chemometrics and Intelligent Laboratory Systems, 14: 225-233. A technical property, namely fastness to light, of a series of mono-azo dyes is modelled by the partial least squares (PLS) method, using different kinds of structure descriptors: molecular descriptors calculated on the basis of partial equalization of orbital electronegativities, ¹³C NMR and ¹H NMR experimental data, and tabulated substituent physico-chemical parameters. Two variable selection approaches in PLS are compared with traditional PLS: a backward variable elimination and a stepwise selection procedure, respectively. For this set of compounds, the best descriptors appear to be the experimental NMR data and the calculated molecular descriptors. The last ones are advantageous since previous synthesis of the compounds is not needed to obtain the value of the descriptors in order to apply the statistical model for prediction purposes. As far as the statistical approach is concerned, both the stepwise and the backward refinements appear to improve the predictive ability of PLS.