A knowledge-driven bi-clustering method for mining noisy datasets

Abstract

Bicluster discovery is an important task in various experimental domains. We propose here a new biclustering system COBIC, which combines graph algorithms with data mining methods to efficiently extract highly relevant and potentially overlapping biclusters. COBIC is based on maximum flow / minimum cut algorithms and is able to take into account background knowledge expressed as a classification, by a weight adaptation mechanism when iteratively extracting dense regions. The proposed approach, when compared on three real datasets (Yeast gene expression datasets) with recent and efficient biclustering algorithms shows very good performances. © 2012 Springer-Verlag.