基于基因表达谱特征分布的SOM聚类算法研究  

Study of SOM clustering algorithm based on features' distribution of gene expression profile

作　　者：程慧杰[1,2] 张国印[1] 何颖[2] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,黑龙江哈尔滨150001 [2]哈尔滨医科大学计算机教研室,黑龙江哈尔滨150081

出　　处：《计算机工程与设计》2011年第7期2463-2466,共4页Computer Engineering and Design

摘　　要：针对SOM算法中欧氏距离无法根据特征的重要性来衡量相似度、易引入无关特征干扰的缺点,提出了一种基于基因表达谱特征分布的SOM聚类算法。算法通过衡量特征对同类基因的凝聚能力和对异类基因的区分能力,对不同的特征赋予不同的权值,将此权值引入到基因数据与神经元的相似度计算中,并利用改进的粒子群优化算法调整获胜神经元及邻接神经元的权值。实验结果表明,该算法有效增强了聚类结果的类内凝聚度和类间区分度,提高了聚类准确率。The Euclidean distance of SOM is impossible to calculate similarity by importance of features and is liable to be lntertered by irrelevant features. A SOM clustering algorithm based on feature＇ s distribution ofgene expression profile is proposed. Feature＇ s ability to aggregate similar genes and distinguish dissimilar genes is evaluated. Different features had different weights and the weights are in- troduced to the similarity computation ofgene data and neurons. The weights of wirming neuron and adjacent neuron are adjusted by an improved PSO algorithm. Experimental results show that the algorithm enhances the cohesion of intra cluster and the discrimination of inter cluster and improve precision of gene clustering.

关 键 词：基因表达谱 自组织映射 粒子群优化 特征权值 基因聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]