Biclustering of Gene Expression Data Using Genetic Algorithm

The biclustering problem of gene expression data deals with finding a subset of genes which exhibit similar expression patterns along a subset of conditions. Most of the current algorithms use a statistically predefined threshold as an input parameter for biclustering. This threshold defines the maximum allowable dissimilarity between the cells of a bicluster and is very hard to determine beforehand. Hence we propose two genetic algorithms that embed greedy algorithm as local search procedure and find the best biclusters independent of this threshold score. We also establish that the HScore of a bicluster under the additive model approximately follows chi-square distribution. We found that these genetic algorithms outperformed other greedy algorithms on yeast and lymphoma datasets.