Abstract
Many researches in molecular genetics area have identified a number of important genes of various types of cancers and the identification of significant biological networks corresponding to gene expression data has been an important issue in understanding underlying biological mechanisms of cells. We integrate phenotype networks, protein networks and apply a greedy Markov blanket search method that efficiently utilizes both gene expression data and protein-protein interaction networks to identify significant networks as well genes for a human disease. We use prostate cancer data as our test domain. In comparison with such statistical methods as t-test and wilcoxon test, our method identifies more prostate cancer-related genes than those reported in published database and literature. We identify disease-related genes with higher precision and at least 1.5 fold higher F-measure. The functional modules involved in the prostate cancer is over-expressed Interleukin-type, insulin-like growth factors and well-known RAS related oncogenes are identified by our method. Cell signaling, immune response, cell cycle and cytokine interactions canonical pathways are also found to be significantly related to prostate cancer. Our proposed methods efficiently utilize gene expression, phenotype and protein networks in identifying the sub-networks and genes that might be related to the disease under interest. Those significant genes and the associated networks may be the subjects to understand the mechanism of prostate cancer. Our method would be more powerful and accurate to integrate the microarray data and the phenotype network for identifying the disease-related genes and networks. Keywords: Prostate Cancer, Microarray data, Protein-protein interaction networks, Markov Blanket search, Phenotype networks