Comparing batch update with randomized update for identifying salient genes applied to cancer gene expression clustering

DNA microarrays usually screen a sufficiently large number of genes, including redundancies. In this paper, we study a neighbour-based method for gene assessment applied to the discovery of interesting clusters in an attempt to understand relations among cancer gene expression data. Using the gene assessment, an adaptive vector space is used for recording the genes’ saliences, where the element in this vector represents the weight of the corresponding gene. We thus compare a batch update strategy to a randomized update strategy to iteratively update vectors in the process of gene assessment. In tests on two benchmark cancer gene expression datasets, the experimental results indicate that our batch update strategy performs better than the randomized update strategy for gene assessment applied to the discovery of interesting clusters.