基于二层特征筛选的HIV-1蛋白酶特异位点预测  

HIV-1 protease cleavage site prediction based on two stage feature selection method

作　　者：袁啸尘[1] 钮冰[1] 尹京苑[2] 

机构地区：[1]上海大学生命科学学院,上海200444 [2]上海大学计算机工程与科学学院,上海200072

出　　处：《生物信息学》2012年第4期274-279,共6页Chinese Journal of Bioinformatics

基　　金：上海市优秀青年教师基金(SHU10022);国家自然科学基金(20973108)

摘　　要：在抗艾滋病治疗中,HIV-1蛋白酶抑制剂发挥着重要作用。对于HIV-1蛋白酶裂解作用位点的研究有助于找到新的治疗靶点。为了对HIV-1蛋白酶特异位点进行预测,本研究用氨基酸索引数据库(Amino Acid Index,AAIndex)中的531个氨基酸物理化学性质参数直接表征肽样本的结构,通过二层特征筛选,最终将4248个表征参数降为57个表征参数。分别采取四种核函数进行HIV-1蛋白酶特异位点的支持向量机(SVM)建模,并通过10折交叉验证及外部测试集方法来验证建模的准确性。结果表明选取NormalizePolyKernel核函数进行SVM建模效果优于其他核函数(PolyKernel、PUK、RBFKernel),所建立的模型对于训练集的10组交叉验证预测准确率达到93.947%,对于外部测试集的预测正确率达到93.684%。The HIV - 1 protease inhibitor plays an important role in the therapy of AIDS. The research on HIV - 1 protease＇ s cleavage site will be useful to found new therapeutic targets. To predict the HIV - 1 protease specific site, we apply Amino Acid Index（AAIndex） ＇ s 531 amino acid＇ s parameter of chemical and physical to present the structure of peptide sample. And based on two stage feature selection method , 57 features are selected from origi- nal 4248 features. By using four kernel function of support vector machine （ SVM）, HIV - 1 protease specific site＇ s model is built. Our research showed the modeling by the kernel function of NormalizePolyKernel had the higher prediction rate than other three kernel function. As a result, the accuracy rate of prediction achieves 93. 947 % and 93. 684% for corss validation test and an independent set test, respectively.

关 键 词：变量筛选 支持向量机 10折交叉验证 预测模型 HIV-1蛋白酶 

分 类 号：O641[理学—物理化学]