Efficient Mining Differential Co-Expression Constant Row Bicluster in Real-Valued Gene Expression Datasets

Biclustering aims to mine a number of co-expressed genes under a set of experimental conditions in gene expression dataset. Recently, differential co-expression biclustering approach has been used to identify class-specific biclusters between two gene expression datasets. However, it cannot handle differential co-expression constant row biclusters efficiently in real-valued datasets. In this paper, we propose an algorithm, DRCluster, to identify Differential co-expression constant Row biCluster in two real-valued gene expression datasets. Firstly, DRCluster infers the differential co-expressed genes from each pair of samples in two real-valued gene expression datasets, and constructs a differential weighted undirected sample-sample relational graph. Secondly, the differential co-expression constant row biclusters are produced in the above differential weighted undirected sample-sample relational graph. We also design several pruning techniques for mining maximal differential co-expression constant row biclusters without candidate maintenance. The experimental results show our algorithm is more efficient than existing one. The performance of DRCluster is evaluated by MSE score and Gene Ontology, the results show our algorithm can find more significant and biological differential biclusters than traditional algorithm.