Biclustering of gene expression microarray data using dynamic deme parallelized genetic algorithm (DdPGA)

Abstract

Biclustering deals with creating a sub-matrix that shows a high similarity across both genes and conditions. Biclustering targets at identifying several biclusters that reveal potential local patterns from a microarray matrix. In this paper, initially sequential evolutionary algorithm (SEBI) is implemented and few drawbacks of the approach were identified. To overcome the drawbacks, parallel strategies such as condition based evolutionary biclustering (CBEB) and coarse grained parallel genetic algorithm (CgPGA) were implemented. To further improve the performance, a new parallel genetic algorithm using dynamic demes strategy is implemented. This method uses global parallelization (master-slave model) with coarse-grained GA with overlapping subpopulation model. The primary objective is to find biclusters with minimum overlapping, large row variance, low mean square residue (MSR) and covering almost every element of expression matrix, thus minimizing the overall fitness value. Sequential EA and condition based EA (CBEB) is implemented but it was observed that both took too much time to meet the stopping criteria. So, to improve the efficiency of the genetic algorithm (GA), Parallel GA has been implemented with dynamic deme strategy to reduce the execution time of GA and find good quality biclusters. DdPGA yielded good quality biclusters and search space could be increased by implementing this strategy. This experiment was implemented on yeast Saccharamyces dataset.