Research On Hybrid Gene Selection Method Based On Clustering

Gene chip can test a biological sample of all the transcriptional activity. Unlikemost experiment tools, microarray can give us a snapshot of the cell on gene level.This modern technology has been widely used in many research fields, includingdiscovering new cancer subset or cancer classification. However, usually geneexpression profile contains thousands of variables while only limited samples andmost of genes are irrelevant with diseases or cancer classification. Gene selection notonly can help us find significant genes but also can save computational complexitycost. In this context, gene selection becomes critical important. The main researchwork is as follows:A hybrid gene selection method based on clustering using feature similarity.According the characteristic of gene expression profile, the original data commonlyhas high redundancy. The redundancy not only make the computational cost veryexpensive but also makes find truly relate genes impossible. In this paper we providea clustering method using feature similarity to solve this problem. Four public geneexpression profile datasets are used to test our method. Experiment results show thatour method can get good classification performance.Considering most gene selection method based on clustering algorithm justselects significant candidate genes with the highest “information exponential” at eachgene clustering. So we implement a maximum cliques search method to find morereliable patterns. The experimental result shows the efficient of method.