Abstract: In bioorganic analysis, abundant structural information can be provided by carbon-13 nuclear magnetic resonance (¹³C NMR) and more and more attentions have recently been paid on its molecular modelling and quantitative prediction which on the basis of the relationship of chemical shift of carbon nuclear magnetic resonance with descriptor variables of chemical structure. By using multiple linear regression (MLR) and latent factor analysis (LFA) techniques, quantitative ¹³C NMR models are achieved to accurately express correlation of ¹³C NMR chemical shifts with five structural parameters and to successfully predict the chemical shift (CS)of any other compounds optimally. First, the history and progress in quantitative structure spectra relationship (QSSR) were critically reviewed, and a set of novel descriptors consisting of 4 elements, called atomic electronegative distance edge vector (AEDV) were first developed by our laboratory and further investigated for the bioactive compounds. Next, MLR and LFA were simply introduced; Matlab and True Basic programs for quantitative molecular modelling (QMM) were designed and written by ourselves. Then, various chemical shifts of ¹³C NMR for 457 different chemically equivalent carbon atoms in 135 natural sugars were estimated and/or predicted with the atomic electronegative distance edge vector (AEDV) with 4 elements and the γ calibration parameter:The correlation coefficients R, roots of mean square error RMS, standard deviation SD, F-statistic value F, and explained variance being n=62, R=0.9910, RMS=1.9602 (SD=1.9762, R²=0.9821, F=502.3294, EV=0.9805); n=79, R=0.9886, RMS=2.5405 (SD=2.5567, R²=0.9773, F=515.6046, EV=0.9757); n=302, R=0.9514, RMS=3.6884 (SD=3.6945, R²=0.9051, F=468.8276, EV=0.9035)and n=14, R=0.5772, RMS=8.8626(SD=9.1972, R²=0.3331, F=0.5828, EV=-0.0837)for the primary, secondary, tertiary and quaternary carbons in all carbohydrates, respectively. Finally, cross validation with leave-one-out (LOO) procedure was made on the QSSR equations for all four types of carbon atoms. The good results were obtained for the first three types of carbon atoms except the quaternary carbon atoms, which indicated that there exists a simply multiple linear relationship between CS and AEDV for primary, secondary, tertiary carbon atoms of sugars.

Key words: Nuclear magnetic resonance, Multiple linear regression, Latent factor analysis, Atomic electronegative distance-edge vector (AEDV), γ calibration, Chemical shift, Quantitative structure-spectra relationship, Sugars, Carbohydrates