Journal of Frontiers of Computer Science and Technology ›› 2011, Vol. 5 ›› Issue (10): 921-931.

• 学术研究 • Previous Articles     Next Articles

The Prediction of MicroRNA in Ensemble of Pairwise Constraints Dimensionality Reduction

WEI Shuang, YANG Ming, XIAO Yuan   

  1. 1. School of Computer Science and Technology, Nanjing Normal University, Nanjing 210046, China 2. Jiangsu Research Center of Information Security & Privacy Technology, Nanjing 210046, China 3. PLA Nanjing Institute of Politics, Nanjing 210003, China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-10-01 Published:2011-10-01

成对约束降维集成下的MicroRNA预测

魏 爽, 杨 明, 肖 袁   

  1. 1. 南京师范大学 计算机科学与技术学院, 南京 210046
    2. 江苏省信息安全保密技术工程研究中心, 南京 210046
    3. 解放军南京政治学院, 南京 210003

Abstract: MicroRNAs (miRNAs) are a class of endogenous RNAs, which play an important role in gene regulation. The prediction of miRNAs will help study and understand their biological functions. The proposed local semi-supervised linear discriminant analysis dimensionality reduction (LSLDA) algorithm, which obtains a low-dimensional space as well as preserves local structure information and discrimination of data, can improve the prediction of miRNAs. Based on the LSLDA algorithm, this paper proposes an ensemble local semi-supervised linear discriminant analysis (En-LSLDA) algorithm by integrating classified results of different numbers of constraints as final results to further improve the prediction of miRNA effectively. Experimental results on miRNA data sets show the effectiveness of the En-LSLDA algorithm. Meanwhile, experimental results on UCI data sets validate the newly proposed algorithm can be applied to other data sets.

Key words: miRNA, pairwise constraints, dimensionality reduction, ensemble, prediction

摘要: MicroRNA(miRNA)是一类在生物体内发挥重要调控作用的非编码小RNA, 对miRNA的预测有助于研究和理解其生物学功能。已经提出的基于成对约束的降维算法(local semi-supervised linear discriminant analysis, LSLDA)在对miRNA降维的同时, 也能保持数据的局部结构信息和判别能力, 可有效改进miRNA的预测性能。因此, 在LSLDA算法基础上, 提出了一种新的集成LSLDA算法(ensemble of local semi-supervised linear discriminant analysis, En-LSLDA)。该算法对不同约束个数下的分类结果进行集成, 以集成结果作为最后的分类结果, 以此进一步改进miRNA的预测性能。miRNA数据集上的实验结果表明, En-LSLDA算法是有效可行的。同时, UCI数据集上的实验结果也验证了新提出的集成方法同样适用于其他数据集。

关键词: miRNA, 成对约束, 降维, 集成, 预测