SVM的参数优化及在耐热酶和常温酶分类中的应用

Parameters Optimization of Support Vector Machine for Discriminating Thermophilic and Mesophilic Enzyme

氨基酸的组成是影响蛋白质耐热性的主要因素之一,所以以20种氨基酸所占比例作为特征向量,利用支持向量机(Support vector machine,SVM)预测蛋白质的耐热性。在比较了几何方法、SVM-KNN和重复训练3种参数优化的方法之后,从中选择了几何方法来优化SVM分类器的参数,并使预测率从85.4%提高到88.2%。从预测率上可知:(1)几何方法优化SVM参数可以有效地提高预测率;(2)氨基酸含量与酶的耐热性之间存在极强的相关性。

It was widely accepted that amino acid composition play vital role on the protein thermostability.In this manuscript,20-amino acid composition in their protein sequence was chosen as the feature vector of SVM and used to predict the protein thermostability by SVM.the accuracy increased from 85.4% to 88.2% by using the geometrical method to optimize SVM parameter.Furthermore,it could be acquired following conclusions：（1） geometrical method is an efficient method to improve the accuracy of SVM parameters;.（2） there is a very close relationship between percentage of amino acid and enzyme thermostability.