基于流形学习的DNA序列数据挖掘方法研究

Research on DNA sequence data mining method based on manifold learning

由于人类DNA序列上单核苷酸具有多态性,DNA序列异常挖掘是后基因组时代的一个重要研究课题。文章在分析现有DNA序列数据挖掘方法的基础上,利用流形学习中不同低维嵌入向量之间向量距离不同的特点,提出了基于流形学习的DNA序列数据挖掘方法(5Dlocally linear embedding,简称5DLLE)。实验结果表明,与隐马尔可夫模型(HMM)和支持向量机(SVM)相比,文中所提出的5DLLE方法在DNA序列数据挖掘方面具有一定优势,不但平均识别率高,而且计算时间相对较少。

For the polymorphisms of single nucleotide on human DNA sequence ,it is important to re-search DNA sequence outlier mining in the post-genome era .In this paper ,based on the analysis of the existing DNA sequence data mining methods ,and by utilizing the characteristic that vector dis-tances between different low-dimensional embedding vectors are different in manifold learning ,a DNA sequence data mining method based on manifold learning ,i .e .5D locally linear embedding （5DLLE） method is proposed .The experimental results show that ,comparing with the hidden Markov model （HMM） and support vector machine（SVM） ,the proposed 5DLLE method possesses some advantages in DNA sequence data mining ,including higher average recognition rate and relatively less calculation time .