Abstract:

In order to study on the features of aromatic composition in wild Chinese chives and predict its chromatographic retention index of volatile constituents molecules, self-designed programs are run in MATLAB to get the molecular shape index and electrotopological state index of volatile constituents of wild Chinese chives. The two indexes are used as molecular structure parameters, and by using multiple stepwise regression method, we screen and optimize the structure parameters 2K、3K、4K、I2 and I6 to establish a QSRR model of chromatographic retention index of volatile constituents of wild Chinese chives. The correlation coefficient of regression equation reaches 0.963. By examining the stability and predictive ability of the model, we find the correlation coefficient r of the inspection is stable at around 0.963. The five structural parameters are used as the input neurons of the artificial neural network and a 521 network architecture is employed. A satisfying neural network model is constructed with the back-propagation(BP) algorithm. The gross correlation coefficient R is 0. 996,which shows a significant correlation. The forecasted values is basically tally with experimental values that we get from the model with only 1.67％ mean relative error. It can be seen that the results of back-propagation network are better than those of multiple linear regression methods.

Key words: quantitative structureretention relationship (QSRR), wild Chinese chive, chromatograph retention index, volatile constituents, artificial neural network(ANN)