Correlative Study On Chromatographic Retention Indices Of Polycyclic Aromatic Sulfur Heterocycles

摘要

Based on the topological theory and MATLAB program, electrotopological state indexes (E_n) were calculated for 114 polycyclic aromatic sulfur heterocycles. A six-element regression model of quantitative structure-retention relationship (QSRR) for retention index (RI) as a function of E_n was constructed using leaps-and-bounds regression (LBR). The traditional correlation coefficient (R), determination coefficient(R~2) and the cross-validation correlation coefficient (Q~2) were 0.995, 0.989 and 0.982 respectively. The model is highly reliable and has good predictive ability. The six structural parameters were used as the input neurons of artificial neural network, and a 6:8:1 network architecture was employed. A satisfied model was constructed with the back-propagation algorithm, the correlation coefficient (R~2) was 0.996. It can be concluded that the prediction results of BP-ANN model are better than MLR-QSRR model.