基于PCA-SVM混合算法的嗜热菌和常温菌网络特征的研究  

Study on thermophile and mesophile characteristics of the network using PCA-SVM hybrid algorithm

作　　者：王文超[1] 丁彦蕊[1] 陈璟[1] 

机构地区：[1]f江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机与应用化学》2012年第6期651-655,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(21001053);中央高校基本科研业务费专项资金资助(JUSRP11126)

摘　　要：通过比较嗜热菌和常温菌代谢网络的特征参数,可以从系统角度确定微生物嗜热性的主要因素。本文首先利用主成分分析法对22个网络特征进行相关性分析,根据特征值、载荷值的大小最终选择了11个主要网络特征;用选出的这11个网络特征组成特征向量,利用支持向量机构建分类器,对嗜热菌和常温菌进行分类,其全局平均预测率为82.93%,对常温菌和嗜热菌的平均预测率分别为87.86%和72.40%。结果表明利用主成分分析法选择的网络特征可以很好的表征嗜热菌和常温菌的耐热性,因此簇大小分布的平均信息等11个网络特征是影响微生物耐热性的关键的代谢网络特征因素。By comparing the metabolic network features of thermophile and mesophile, it can determine the main factors of thermophilie from the system perspective. In this paper, first, using principal component analysis of 22 network features in correlation analysis. According to the eigenvales and loading values 11 main features of the networks are chosen. Formed the eigenvectors by these network characteristics, using SVM constructed the classifier,classified thermophile and mesophile, the average predictive of global is 82.93%, the average predictive of thermophile and mesophilc are 87.86% and 72.40%. The results show that the network features chosen by the principal component analysis can well characterize the heat resistance of thermophile and mesophile, so average information oncliquc size distribution etc. 11 network features is the key factors of metabolic networks on the microbial heat resistance.

关 键 词：主成分分析法 SVM算法 网格搜索 网络特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] Q5[自动化与计算机技术—计算机科学与技术]