A New Tissue Clustering Algorithm Based On Gene Clustering With SOM

Along with the research and extensive application of DNA chip technology, gene expression data analysis have become a hotspot in life science field. Tissue clustering, the process of grouping related samples according to gene expression levels, is very useful to the research of gene unknown functions, and is one of the most important foundation of microarray study. Microarray data has such a feature that it has a few tissue samples but each corresponds to expression levels of a huge number of genes, in which a lot interfere the grouping structure of the data and bother the feasible clustering of the tissues. In this paper, we present a tissue clustering method based on the fact that only a few number of genes, i.e., the contributive genes, are first selected from the huge number of genes, and the tissue samples are clustered with only the expression levels of these contributive genes. In our method, the contributive genes are selected via a circular-structured SOM, in which the more distant some two genes in the SOM, the more independent they function for clustering. The sensitivity of a gene is defined for the measurement of the contribution of the gene for gene selection from the output of the SOM, and both k-means method and SOM method are used for tissue clustering. Our experimental result on real microarray shows the great effectiveness and efficiency of the method presented in the paper. On this basis further study is made on the features of Pulse Coupled Neural Network. Then a clustering algorithm based on a PCNN is presented, which is of high parallelism and can get the clustering outcome faster and more exactly than other methods. The experimental results prove its great advantage over others.