Quantitative structure spectroscopy relationships of carbon-13 nuclear magnetic resonance chemical shifts of steroids

Quantitative structure spectroscopy relationships (QSSRs) are systematically studied for carbon-13 nuclear magnetic resonance (13C NMR) spectroscopic simulation of steroid compounds. Both the atomic electronegativity interaction vector (AEIV) and the atomic hybridization state index (AHSI) are used for the expression of local chemical microenvironment and atomic hybridization state of 4434 resonance carbon atoms in 203 steroid molecules. A multiple linear regression (MLR) model is built after screening some insignificant parameters with the stepwise multiple regression (SMR) technique. Correlation coefficients of the developed model are Rcum2=0.9341 and QLOO2=0.9336 for classical estimation of molecular modeling and the cross-validation with leave-one-out (LOO) procedures, respectively, primarily indicating that the MLR model has good modeling stability and prediction ability. Furthermore, the superior performance of the MLR model is tested by the leave-33%-out (L33%O) cross-validation method, where the mean correlation coefficients of three test sets are Q2 = 0.9310 and Qext2=0.9196 for both internal and external sets. In conclusion, AEIV and AHSI descriptors can be used for estimating and predicting 13C NMR chemical shifts of steroids.