基于SVM和优化特征集的MicroRNA靶标预测  

MicroRNA Target Predicition Based on SVM and the Optimized Feature Set

作　　者：王宝文[1] 齐晓阳 王常武[1] 刘文远[1] 司亚利[1] 

机构地区：[1]燕山大学信息科学与工程学院,秦皇岛066004

出　　处：《生物医学工程学杂志》2013年第6期1213-1218,共6页Journal of Biomedical Engineering

摘　　要：MicroRNAs(miRNAs)是一类长约22个核苷酸序列(nt)的非编码单链小RNA,通过与靶标mRNA 3′-UTR完全或不完全碱基互补匹配,在后转录时期发挥重要的基因调控功能。正确识别miRNA阳性靶标已成为研究miRNA功能的瓶颈。针对miRNA靶标预测的高维、非线性小样本数据集,本文基于υ-SVM,提出冗余特征剔除算法,并将模式识别和特征选择融合在一起,通过冗余特征剔除算法优化特征集,构造表征miRNA与靶标作用模型的最佳特征组合,先验参数γ(0<υ≤1)控制数据集压缩比例,选取有区分力度的支持向量集,进而构建最佳预测性能的miRNA靶标分类器模型。采用独立测试集对miRNA靶标预测模型进行无偏差性能评估。试验表明,文中靶标预测的分类识别和泛化性能比算法miTarget、NBmiRTar及TargetMiner等更佳。MicroRNA （miRNA） is a family of endogenous single-stranded RNA about 22 nucleotides in length. Through targeting 3UTR of message RNA（mRNA）, they play important roles in post-transcriptional regulatory functions. For further research of miRNA function, the identification of more miRNA positive targets is needed ur- gently. Aiming at the high-dimensional small sample data sets in miRNA target prediction, an algorithm of elimina- ting redundant features is proposed based on u-SVM in this paper, and classification and features selection are also fused. The algorithm of eliminating redundant features optimizes the combination of features, and then constructs the best features combination which can represent miRNA and targets interaction model. The prior parameter u（0~u~ 1） controls the compression proportion of data set and selects more distinguishing support vectors. Finally, the clas- sifier model of miRNA target prediction is built. The unbiased assessment of the classifier is achieved with acom pletely independent test dataset. Experiment results indicated that in both classification recognition and generalization performance of miRNA targets predicition,this model was superior to the present machine learning algorithms such as miTarget, NBmiRTar and TargetMiner, etc.

关 键 词：MICRORNA 支持向量机 靶标预测 特征选择 分类器模型 

分 类 号：Q78[生物学—分子生物学] TP181[自动化与计算机技术—控制理论与控制工程]