A Biclustering Method to Discover Co-regulated Genes Using Diverse Gene Expression Datasets

Abstract. We propose a two-step biclustering approach to mine co-regulation patterns of a given reference gene to discover other genes that function in a common biological process. Currently, several successful methods utilize Pearson Correlation Coefficient (PCC) based gene expression analysis across all samples in datasets. However, microarray datasets are fraught with spurious samples or samples of diverse origin, and many genes/proteins that function in the same biological pathway may be missed. The novel PCC based biclustering algorithm introduced in this paper identifies subsets of genes with high correlation by stringently filtering the data and reducing false negatives due to spurious or unrelated samples in a dataset. Then, correlation information extracted from resulting biclusters are synthesized. We applied our method using the breast cancer associated tumor suppressors, BRCA1 and BRCA2, as the reference proteins to reveal genes and proteins important in the complex process of breast tumor formation. Experiments on 20 very large datasets showed that the top-ranked genes were remarkably enriched for genes that regulate the mitotic spindle and cytokinesis. The results imply that BRCA1 and BRCA2 proteins, which are considered to be DNA repair factors, have critical function regarding the mitotic spindle as well. Initial biological verification reveal that this identified factor function to control both centrosome dynamics, and also, surprisingly, DNA repair. Thus, this biclustering approach is successful at identifying proteins with highly related function from extremely complex datasets, and permits novel insights into gene function.