新型适用于基因表达数据的模型聚类方法  

A new model clustering method for gene expression data

作　　者：宋东奇[1] 宋余庆[1] 刘哲[1] 凌青华[1] 

机构地区：[1]江苏大学计算机科学与通信工程学院,江苏镇江212013

出　　处：《计算机与应用化学》2015年第1期71-74,共4页Computers and Applied Chemistry

基　　金：江苏省自然科学基金(BK20130529);高等学校博士学科点专项科研基金资助课题(20113227110010);江苏省博士后科研资助计划项目(1202037C);中国博士后科学基金(2013M541616)

摘　　要：针对EM算法中的初始类的数目很难决定,在迭代中经常产生部分最优的情况,将K-means算法与基于EM的聚类方法相结合,提出了一个新的适用于基因表达数据的模型聚类方法。新的聚类方法,首先利用K-means算法具有全局性、效率高的优点,快速得到聚类的起始类的划分,将其设置为高斯混合模型的初始参数值,进一步采用EM方法进行聚类,得到最优聚类结果。通过2次对真实数据集的实验测试,将新的算法分别与K均值算法和EM算法进行了比较。实验结果表明,新算法是一种有效的聚类方法,聚类结果的准确度得到了提高。The EM algorithm was difficult to determine the number of clusters and often produced the local optimal condition in the process of iteration. By combining k-means algorithm and EM algorithm, a new model clustering method which applies to gene expression data was put forward. Firstly, the overall and highly active advantages of the k-means clustering algorithm were used to get the initial class of the clustering quickly, which was served for the initial value of the Ganssian model and then clustering by EM algorithm to get the optimal clustering results. Two experiments were performed by the real datasets and compared the new algorithm with the k-means algorithm and EM algorithm. Experimental results show the mixture algorithm is an effective clustering algorithm for gene expression data and the discrimination of clustering get better.

关 键 词：基因表达数据 K-MEANS算法 EM聚类算法 生物信息学 

分 类 号：TQ015.9[化学工程] TP391.9[自动化与计算机技术—计算机应用技术]