The Research Of Improved Multi-objective Genetic Algorithms For Biclustering On Gene Expression Data

Gene expression data reflects the expression levels of thousands of genes under different experimental conditions,during the investigation process of the gene expression data,we can obtain significant biological information by discovering the hidden similar expression patterns.Considering that the gene expression data usually contains a large amount of data,as well as some noise.Therefore,how to extract useful biological information from the gene expression data becomes a worthwhile challenge.According to the research in the bioinformatics,we find that the genetic patterns usually consist of the expression levels of a subset of genes under a given subset of experimental conditions.Such a fact is consistent with the phenomenon in gene regulation,specifically speaking,the synergistic genes which show a similar genetic pattern are collaborating only in a certain subset of conditions.Therefore,we apply the biclusteirng algorithm to mine useful biological information in the gene expression data,the biclusteirng algorithm can simultaneously cluster the rows and columns of the gene expression data,resulting in the discovery of the local similar expression patterns.When we detect the biclusters in the gene expression data,we desire to find biclusters with higher quality and larger size,however,these two objectives conflict with each other.Consequently,we design a biclustering algorithm based on multiple objective genetic algorithm.Firstly,we propose an improved multi-objective genetic algorithm.Compared to the traditional multi-objective genetic algorithm,we design a novel population initialization strategy and an innovative selection operator.By comparing our algorithm with traditional multi-objective genetic algorithm under several test problems,we verified the effectiveness of our algorithm.On the other hand,based on the improved multi-objective genetic algorithm,we propose an innovative biclustering algorithm.Firstly,apart from the bicluster population,we incorporate the population of the rows or columns.The population consist of rows or columns can help us to evaluate the contribution of each row or column in the biclusters.As the population of the rows or columns will evolve to the rows or columns that have a positive contribution to the quality of the bicluster,the combination of the population of biclusters and the population of rows or columns can help us to discover the meaningful subspace in finding the global optimal solutions.During the evolutionary process of the bicluster population and the rows or columns population,the cuckoo search algorithm is used for the searching of the optimal solutions in the rows or columns population,the proposed improved multi-objective genetic algorithm is utilized in the evolutionary process of the bicluster population.After that,we combine the evolved rows or columns population and the preliminary evolved bicluster population,the preliminary biclusters can be further evolved and the fitness of the rows or columns population can be updated.By comparing with several traditional biclustering algorithms on multiple synthetic datasets and real datasets,the proposed biclustering algorithm show a better performance.