Bi-clustering of microarray data using a symmetry-based multi-objective optimization framework

Abstract

High-throughput technologies, like DNA microarray, help in simultaneous monitoring of the expression levels of thousands of genes during important biological processes and over the collection of experimental conditions. Automatically uncovering functionally related genes is a basic building block to solve various problems related to functional genomics. But sometimes a subset of genes may not be similar with respect to all the conditions present in the dataset; thus, bi-clustering concept becomes popular where different subsets of genes and the corresponding subsets of conditions with respect to which genes are most similar are automatically identified. In the current study, we have posed this problem in the multi-objective optimization (MOO) framework where different bi-cluster quality measures are optimized simultaneously. The search potentiality of a simulated annealing-based MOO technique, AMOSA, is used for the simultaneous optimization of these measures. A case study on the suitability of different distance measures in solving the bi-clustering problem is also conducted. The competency of the proposed multi-objective-based bi-clustering approach is shown for three benchmark datasets. The obtained results are further validated using statistical and biological significance tests.