结构指数预测古井贡酒风味成分的色谱保留指数

为建立古井贡酒风味成分保留指数的定量结构-保留相关性（QSRR）模型，计算了古井贡酒风味成分的分子连接性指数、分子形状指数、电性拓扑状态指数和电性距离矢量，优化筛选了分子连接性指数的0X、1X、3X和5Xc，分子形状指数的K1、K2和K3，电性拓扑状态指数的E1和电性距离矢量的m1，将这9种指数与古井贡酒风味成分的色谱保留指数进行回归分析，以这9种分子结构指数作为反向传播（BP）神经网络的输入参数，保留指数作为输出参数，采用9∶13∶1的网络结构，构建了BP神经网络预测模型，总的相关系数rt为0.996 6，计算的预测值与文献值较为吻合，平均相对误差为1.88%。结果表明，模型具有良好的预测保留指数的能力，从构建的模型可知，甲基等取代基数量及所处位置是影响古井贡酒风味成分色谱保留指数大小的主要因素。

Prediction of chromatographic retention indexes of flavor components in Gujinggong Baijiu by structure indexes

In order to establish quantitative structure-retention relationship (QSRR) model of chromatographic retention indexes of flavor components in Gujinggong Baijiu (Chinese liquor), the molecular connectivity indexes, molecular shape indexes, electrotopological state indexes and the electronegativity distance vectors of flavor components in Gujinggong Baijiu were calculated. The molecular connectivity indexes (0X, 1X, 3X and 5Xc), molecular shape indexes (K1, K2 and K3), electrotopological state index (E1) and electronegativity distance vector (m1) were optimized and selected. Using the nine molecular structure indexes as input parameters of back propagation (BP) neural network, retention index as output parameters, the 9∶13∶1 network structure was adopted and the prediction model of BP neural network was established. The total correlation coefficient rt was 0.996 6, the predicted values was consistent with the literature values, and the average relative error was 1.88%. The results showed that the model had good predictive ability to predict the chromatographic retention indexes of flavor components. According to the established model, the number and connection location of substituent group (methyl etc.) were the main factors affecting the chromatographic retention indexes of flavor components in Gujinggong Baijiu.