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.
出处
《计算机工程与设计》
CSCD
北大核心
2011年第7期2463-2466,共4页
Computer Engineering and Design
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
gene expression profile
self organizing map
particle swarm optimization
weight of features
gene clustering