Complete sets of descriptors for the prediction of 13C NMR chemical shifts of quinoline derivatives

Nowadays, there is no criterion to judge whether a descriptor subset has redundant structure information or not, although some criteria can be used to validate the quality of quantitative structure-activity relationship (QSAR) models. This paper reports complete sets of descriptors used for selecting the subset of descriptors for QSAR. Four complete sets of descriptors calculated with B3LYP/6-31G(d) and PBE1PBE/6-311G(2d,2p) approaches were used to develop four QSAR models for 269 C-13 nuclear magnetic resonance (NMR) chemical shifts (delta(C) parameters) of carbon atoms in 26 quinoline derivatives. Four QSAR models for delta(C), parameters were constructed with support vector machine (SVM) algorithm by applying genetic algorithm (GA) to optimize SVM parameters C and gamma. The four SVM models have root-mean-square (RMS) error range of 2.0 ppm to 2.7 ppm. Compared with previous QSAR models for C-13 NMR chemical shifts, the prediction results are accurate, which suggest that applying complete sets of descriptors for QSAR models is successful.