基于混合核和加权SVM的microRNA前体识别算法

MicroRNA是一种单链RNA小分子,是由具有发夹结构的、更长的单链RNA前体经加工后生成.相比microRNA序列本身而言,其前体序列和二级结构隐含了更多的可识别特征与信息.因此可利用加权Levenshtein距离,结合其前体序列和二级结构构造一个指数核函数.结合SVM构造识别模型,鉴别真假前体.在用5折叠法得到最佳识别模型后,对人类数据进行测试.实验结果显示,新方法表现出了较好的识别精度,和较高的敏感性与特异性.

Detecting True and False microRNA Precursors Using Levenshtein Mixed Kernel and Weighted Support Vector Machine

icroRNAs (miRNAs) areRNA molecules that are synthesized from a longer precursor(pre-miRNA) forming a long hairpin structure.MicroRNAs play an important regulatory functions in eukaryotic gene expression through mRNA degradation or translation inhibition.In this thesis we present a kernel-based machine learning method for classifying microRNA by learning from secondary structure and sequence together. The kernel function measures the similarity between a pair of inputs, and defines an inner product in an implicit feature space for the SVM optimization problem. The string kernel consisting of two exponential kernels is built with pre-miRNA sequences and their secondary structure, in which distance measure between two vectors is replaced by weighted Levenshtein distance (WLD) between two sequences.