A time-series biclustering algorithm for revealing co-regulated genes

Abstract

Although existing bicluster algorithms claimed to be able to discover co-regulated genes under a subset of given experiment conditions, they ignore the inherent sequential relationship between crucial time points and thus are not applicable to analyze time-series gene expression data. A simple and effective deletion-based algorithm, using the mean squared residue score as a measure, was developed to bicluster time-series gene expression data. While enforcing a threshold value for the score, the algorithm alternately eliminates genes and time points according to their correlation to the bicluster. To ensure the time locality, only the starting and ending points in the time interval are eligible for deletion. As a result, the number of genes and the length of time interval are simultaneously maximized. Our experimental results shown that the proposed method is capable of identifying co-regulated genes characterized by partial time-course data that previous methods failed to discover.