木聚糖酶氨基酸组成与其最适pH的神经网络模型

A Model for Amino Acid Composition and Optimum pH in G/11 Xylanase Based on Neural Networks

籍均匀设计(UD)方法,构建了G/11家族木聚糖酶氨基酸组成和最适pH的神经网络(NNs)模型。当学习速率为0·09、动态参数为0·4、Sigmoid参数为0·98,隐含层结点数为10时,该模型对最适pH的拟合和预测平均绝对百分比误差可分别达到3·02%和4·06%,均方根误差均为0·19个pH单位,平均绝对误差分别为0·11和0·19个pH单位。该结果比文献报道的用逐步回归方法好。

In this paper, a prediction model for amino acid composition and optimum pH of xylanase in G/11 family was established in terms of an artificial neural networks based on uniform design. Results showed that the calculated and predicted pHs fitted the optimum pHs of xylanase very well and the MAPEs (Mean mean Absolute Percent Error) were 3.02% and 4.06%, the MSEs (Mean Square Error) were 0.19 and 0.19 pH unit, the MAE (Mean Absolute Error) were 0.11 and 0.19 pH unit,respectively. It was better in fittings and predictions compared with the reported model based on stepwise regression.