关键词: 聚类, 样本关系网络, 结构特征, 关系特征

Abstract:

A new clustering method based on the relationship between patterns is proposed. The relationship between patterns is obtained from gene expression data through the pearson correlation coefficient, which is denoted by a network, the relation feature between patterns is extracted by discovering the structure feature of the network, and clustering is performed in the relation feature space. The proposed method uncovers the dissimilarity between patterns belonging to different classes more effectively, and the dimensionality of the clustering space is so low than there is no need to reduce dimensions. The comparison of the method with the conventional ones shows that the method can obtain a much higher clustering efficiency than other methods and it can lead to a better efficiency even for those data with promiscuous distribution.

Key words: clustering, pattern relation network, structure feature, relation feature