A repeated local search algorithm for biclustering of gene expression data

Given a gene expression data matrix where each cell is the expression level of a gene under a certain condition, biclustering is the problem of searching for a subset of genes that coregulate and coexpress only under a subset of conditions. The traditional clustering algorithms cannot be applied for biclustering as one cannot measure the similarity between genes (or rows) and conditions (or columns) by normal geometric similarities. Identifying a network of collaborating genes and a subset of experimental conditions which activate the specific network is a crucial part of the problem. In this paper, the BIClustering problem is solved through a REpeated Local Search algorithm, called BICRELS. The experiments on real datasets show that our algorithm is not only fast but it also significantly outperforms other state-of-the-art algorithms.