采用BP算法的多层感知机模型的蛋白识别

采用误差反传(BP)算法的多层感知机模型,对嗜热蛋白和常温蛋白进行模式识别.通过增加训练数据及多种检验方法检验模型稳定性及泛化能力,探讨蛋白分子大小对识别效果影响.结果表明,当动态参数为0.2,学习速率为0.5,隐含层节点数为11时,该模型在自一致性检验、交叉验证和独立样本测试3种检验方法中的识别精度分别为91.5%,88.2%和92.1%,其表现优于一些常见的模式识别算法,且具有良好的稳定性及泛化能力.此外,对于较大的或者中等大小蛋白质分子,其识别的精度都较高;而对于较小的蛋白分子,其识别效果较差.

Application of a BP Algorithm Based Multi-Layer Perceptron Model to Discriminate Thermophilic and Mesophilic Proteins

In this paper,a back-propagation(BP) algorithm based multi-layer perceptron model was proposed to discriminate thermophilic and mesophilic proteins.When the momentum parameter,learning rate and the number of the hidden layer nodes were 0.2,0.5 and 11,respectively,the model had the best performance.The success rate for self-consistency check,cross-validation and independent test with other dataset was 91.5%,88.2% and 92.1%,respectively.It outperformed other pattern recognition methods such as K-nearest neigh...