The Research On Significant Genes Selection Based On Ant Clustering

With the completion of human genome project, life sciences have entered a newera. Biologists, using microarray technologies, can analysis thousands of geneexpression values at one time. So they can have an overall understanding of the tissuecell. However, this technology is highly automated, large-scale and miniaturized,gene expression profiles commonly have thousands of gene expression values butonly limited tissue samples, for some reasons. Consequently, how to analysis andprocess these mass of data and find meaningful gene subset, for disease researchmentand treatment, has became a key factor in this domain.Ant colony clustering is based on tomb sweeping principle. Firstly, data objectsare randomly distributed on a2D board. And then ants try to make decisions,consisting pick up or drop objects, based on objects distance in the same cell. Repeatthese two steps until reach the maximum number of iterations. In the past few years,many intelligence swarm algorithms have been implemented into gene selectionmethod. And some of them have got good results.In this paper, we propose an ant clustering algorithm based on grid. Firstly, dataobjects are randomly distributed on a2D board, which can be considered as a matrixC of m m cells. Then an ant colony is used in clustering data.In recent years, domestic and foreign researchers have proposed various geneselection methods. However problems, such as gene collinearity, lack of considerationfor combination genes, and work complexity, are not thoroughly examined andworked out. To solve these problems, we present a new hybrid method, based on antcolony clustering algorithm, for gene selection. We first use filter method to rank thegenes in terms of their expression difference, and then select ‘important’ genes withhigh ‘score’. Afterwards, an ant colony is used in clustering gene expression data.A support vector machine is applied to validate the classification performance ofcandidate genes. The experimental results on four datasets demonstrate theeffectiveness of our method in addressing the problems.