Application of reactive GRASP to the biclustering of gene expression data

Abstract

A bicluster in gene expression dataset is a subset of genes that exhibit similar expression patterns through a subset of conditions. In this work biclusters are identified in two steps. In the first step high quality bicluster seeds are generated using KMeans clustering algorithm. These seeds are then enlarged using Reactive Greedy Randomized Adaptive Search Procedure (RGRASP) which is a multi-start metaheuristic method in which there are two phases, construction and local search. The objective here is to identify biclusters of maximum size with MSR lower than a given threshold. Experiments are conducted on both Yeast and Human Lymphoma datasets. The Experimental results on the benchmark datasets demonstrate that RGRASP is capable of identifying high quality biclusters compared to many of the already existing biclustering algorithms. Compared to the already existing algorithm based on the same RGRASP metaheuristics biclusters with larger size and lower mean squared residue are obtained using this algorithm in Yeast dataset. Moreover in this study the RGRASP is applied for the first time to find biclusters from the Human Lymphoma dataset.