The Research Of Genetic Algorithms For Biclustering On Gene Expression Data

The analysis of gene expression data are significantly important to the study of gene regulation mechanisms and drug treatment of tumor diseases.With the development of gene chip technology,the data volume of the gene expression data is also growing exponentially.Therefore,how to effectively mine useful biological information from the massive gene expression data is a worthwhile challenge.The biclustering algorithm provides an effective method for the analysis of gene expression data.The biclustering algorithm can find gene sets with similar expression patterns under specific experimental conditions,which breaks through the limitations of traditional clustering methods.Therefore,the biclustering algorithm has been a hotspot in the research of gene expression data analysis currently.Among the biclustering algorithms,genetic algorithms have been widely applied to the biclustering problems because their excellent global search ability.Most of the traditional biclustering algorithms based on genetic algorithms consider the quality of a bicluster as a whole without considering the contribution of each row or each column to a specific bicluster.Therefore,most of the biclustering algorithms based on genetic algorithms only simply consider the evolution of a single population with bicluster as individuals.With the rapid growth of gene expression data,the size of the search space is becoming extremely huge.With such huge search space,the evolution search of a single population is more likely to converge on local optimal optima and cannot find the global optimal solution effectively.With this consideration,in this paper,apart from the traditional population of biclusters individual,we innovatively propose a new population with rows and columns as individuals.The corresponding coding schemes and the fitness functions for the above two population are also proposed.Apart from this,the traditional population of bicluster and the new population with rows and columns as individuals proposed in this paper are two different types of populations and should correspond to two different evolutionary learning phases.Therefore,for these two different populations,this paper designs a bi-phase evolutionary architecture where two populations evolve in two phases.The mating process between the two populations facilitates the exchange of the information among the populations,this leads to the promotion of the evolutionary learning of the two populations and the obtaining of the better biclustering results.In this paper,based on the new bi-phase evolutionary architecture,two bi-phase evolutionary biclustering algorithms based on a classical single-objective genetic algorithm and a multi-objective genetic algorithm are proposed separately for gene expression data analysis.In order to verify the rationality and validity of the proposed biclustering algorithms,this paper designs a number of comparative experiments from the two aspects of the synthetic dataset and the real dataset.The experimental results show that the two proposed algorithms have better performance than other algorithms in the synthetic dataset and the real dataset.