Abstract: With the advent of gene expression data, it has shown great promise to explore cancer pathogenesis at the level of molecular biology. Exploring the key genes related with carcinoma is of great importance to the cancer diagnosis and treatment During the past few decades, many approaches were successfully proposed to select the key genes related with carcinoma. However, the gene sets selected by different methods are not always consistent with each other, this will cause difficulty to biological interpretation and practical application. A novel approach based on integrated informalion was proposed for the selection of key genes from gene expression data in this study, the three kinds of information,namely information gain, global discriminate ability and local discriminate ability were derived firstly, then they were weighted and integrated into a score to characterize the discriminate ability of each gene, and the gene set with the highs st classification accuracy was selected as key gene set. The experimental results showed that our approach can get better performance than those based on single information.

Key words: Gene expression data, Gene selection, Information gain, Global reparability, Local reparability, Integrated information