Biclustering of Coherent Time Series in Microarray Data

After recognition of sequences of different genomes, the next logical step is the discovery of their function and regulation. To classify genes in the laboratory, factors such as the behavior of genes, gene expression control and protein interactions have been reviewed. It is expected that genes with similar regulation mechanisms have the same expression patterns.

In this paper, we introduce a special way of clustering, called biclustering, for microarray data obtained from multiple sclerosis (MS) patients. From a biological perspective, gene regulatory modules consist of genes that have similar behaviors at different points of time under several conditions. By identifying these modules, the recognition of the regulatory mechanisms that are the common causes of genes behaviors might be conceivable.

We used a modified format of iterative signature algorithm (ISA) to extract co-expressed gene profiles from microarray data. The combination of K-nearest neighbor (KNN) algorithm and ISA provides a helpful algorithm which results in an outstanding and optimum way to obtain similar genes in microarray data.

The algorithm was performed on a synthetic as well as a real database (MS patients’ data), and showed a pronounced difference between the extracted modules in contrast to ISA. Although we showed our method’s efficiency over synthetic and MS data, it will be usable for any other kinds of data. In other words, our method is based on a series of logical and statistical methods rather than data-based methods.