BP和概率神经网络预测微生物热稳定性的比较  

Comparison of the BP and Probabilistic Neural Network Used in Prediction of Microorganism Thermostability

作　　者：丁彦蕊[1] 徐星宇[1] 须文波[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《江南大学学报（自然科学版）》2012年第6期637-641,共5页Joural of Jiangnan University (Natural Science Edition)　

基　　金：国家自然科学基金项目(21001053);中央高校基本科研业务专项基金项目(JUSRP11126)

摘　　要：微生物的热稳定性与代谢网络的拓扑特征之间存在着密切的关系。分析了460个微生物的22个代谢网络拓扑特征,分别利用BP神经网络和概率神经网络建立分类模型,以比较两种算法预测微生物热稳定性的效果。通过分析隐层数目和扩展常数对分类效果的影响发现,相对于BP神经网络,简单易用、稳定性好的概率神经网络更适合于从代谢网络特征角度预测微生物的热稳定性。当耐热菌与常温菌数比例为1∶1、扩展常数为0.09时,概率神经网络对常温菌、耐热菌的预测率分别为72.83%和82.61%。预测率也表明,聚集系数、介数等代谢网络拓扑特征影响着微生物的热稳定性。There is a significant relationship between microorganism thermostability and metabolic network topology. In this paper, we firstly analyzed 22 metabolic network parameters of 460 microorganisms, and then, we built the classificafion models using BP Neural Network and Probabilistic Neural Network respectively to compare the efficiency of predicting the microorganism thermostability. After studying the influence of the hidden layer number and spread constant on the classifi- cation results, we found Probabilistic Neural Network is suitable for predicting the microorganism thermostability by using metabolic network parameters because of its easiness of use and high stability comparing with BP Neural Network. When the ratio of thermophilic microorganism number ： mesophilic microorganism number is 1 ： 1, and the expand constant is 0.09, the prediction accuracies of mesophlic and thermophilic microorganism are 72.83% and 82.61% respectively. The prediction accuracies also show that the metabolic network parameters including clustering coefficient and betweeness can influence the microorganism thermostability.

关 键 词：BP神经网络 概率神经网络 代谢网络拓扑特征 热稳定性预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]