Bi-clustering Gene Expression Data Using Co-similarity

We propose a new framework for hi-clustering gene expression data that is based on the notion of co-similarity between genes and samples. Our work is based on a co-similarity based framework that iteratively learns similarity between rows using similarity between columns and vice-versa in a matrix. The underlying concept. which is usually referred to as bi-clustering in the domain of bioinformatics, aims to find groupings of the feature set that exhibit similar behavior across sample subsets. The algorithm has previously been shown to work well for document clustering in a sparse matrix representation. We propose a variation of the method suited for analyzing data that is represented as a dense matrix and is non-homogenous as is the case in gene expression. Our experiments show that, with the proposed variations, the method is well suited for finding bi-clusters with high degree of homogeneity and we provide empirical results on real world cancer datasets.