一种考虑基因间相互关系的投影聚类算法  

Projected Clustering Algorithm with the Consideration of the Correlation Among Genes

作　　者：赵宇海[1,2] 王国仁[1,2] 于长永[1,2] 毛克明[1,2] 

机构地区：[1]医学影像计算教育部重点实验室东北大学,辽宁沈阳110004 [2]东北大学信息科学与工程学院,辽宁沈阳110004

出　　处：《小型微型计算机系统》2009年第7期1256-1262,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60803026;60873011;60773219)资助;教育部博士点(新教师)基金项目(20070145112)资助;教育部重大培育项目(706016)资助;国家"八六三"计划项目(2007AA01Z192)资助

摘　　要：针对现有基因表达数据投影聚类算法假定基因相互独立,根据每个基因的独立区分度选择相关投影空间的不足,提出了根据基因间相互关系进行投影聚类的算法MOLION.通过将基因表达数据转换为序列数据,基于设定的用户偏好函数,采用分界判定法对样本穷举树进行快速地深度优先遍历,同时应用了高效的削减和优化策略.几个真实微阵列数据集上的实验证实了提出的算法具有较高的效率和预测准确性,为考察疾病表型的形成原因提供了一个新视角.Most of the existing methods perform the projected clustering for gene expression data under the assumption of independence among genes and choose relevant projected space according to individual discriminative score of single gene. In this paper, a novel projected clustering algorithm, namely MOLION, with the consideration of the correlation among genes is proposed to address this issue. Based on the sequential data transformed and a given user-specific favourite function, MOLION employs the branch and bound paradigm to conduct a quick depth-first traverse on the sample enumeration tree while adopting efficient pruning rules and optimization strategy. The experiments conducted on several real Microarray datasets prove the proposed MOLION is of better efficiency and higher prediction accuracy, which provides a fire-new insight into the pathogenesis problem.

关 键 词：基因表达数据 投影聚类 基因序列 数据挖掘 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]