基于SVM方法构建细菌sRNA靶标预测模型  

Construction of a model for prediction of bacterial sRNA targets using support vector machines

作　　者：赵雅琳[1] 李华[1] 侯妍妍[1] 查磊[1] 曹源[1] 王立贵[1] 应晓敏[1] 李伍举[1] 

机构地区：[1]军事医学科学院基础医学研究所计算生物学中心,北京100850

出　　处：《军事医学科学院院刊》2008年第4期375-378,共4页Bulletin of the Academy of Military Medical Sciences

基　　金：国家"863"高技术项目(2006AA02Z323);国家自然科学基金项目(30500105;30470411)

摘　　要：目的:为实验方法鉴定细菌sRNA靶标和研究sRNA功能提供生物信息学支持。方法:首先以实验证实的132个sRNA与靶标相互作用数据为训练集,其中包含46个阳性数据和86个阴性数据;其次,以实验证实的22个阳性数据和随机生成的1 700个阴性数据为测试集;最后以RNA二级结构谱等特征为变量,运用支持向量机(SVM)方法构建sRNA靶标预测数学模型。结果和结论:构建的模型对训练集的敏感性和特异性均为100%,对测试集的敏感性和特异性分别为72.73%和80.65%。所构建的数学模型为实验发现sRNA靶标提供了生物信息学支持。Objective： To provide bioinformatics support for experimental identification of bacterial sRNA targets and for the study of sRNA functions. Methods： To construct a model for prediction of bacterial sRNA targets, 132 sRNA-mRNA interactions verified by experiments were collected first as the training dataset, which contained 46 positive samples and 86 negative samples. Then, 22 sRNA-mRNA interactions verified by experiments as the positive test dataset and 1700 randomly-generated sRNA-mRNA interactions as the negative test dataset were selected. Finally, support vector machines （SVM） were used to construct the model with the profile of sRNA-mRNA secondary structure as the features. Results and Conclusion ：The model＇s sensitivity and specificity were 100.00% and 100.00% for the training data, and 72.73% and 80.65% for the test dataset, respectively. Therefore, the model provides bioinformatics support for experimental identification of sRNA targets.

关 键 词：SRNA 靶标 预测 机器学习 SVM 

分 类 号：Q78[生物学—分子生物学]