Research On Biclustering Methods Based On Intelligent Optimization Algorithm

Gene controls the trait and activity of a organism.And gene expression data is a main type of data which includes the expression of gene activity.At the same time,the regulation information of genes on organisms is also hided in gene expression data.Many biological data could be dug out by researchers through the analysis of these expression data,and these information have contributions to the understanding of the life activities of organisms.The expression level of the gene is reflected by the gene expression data which acquired by a certain technology,and the research on the analysis method of the gene expression data has always been the focus of researcher.At present,the commonly used technique to mine valuable biological information from gene expression matrix is clustering,but those traditional clustering methods can only mine the similar expression patterns of some genes under all conditions.However,in practice,some genes have similar expression patterns under some conditions,thus many scholars put forward the concept of biclustering.Especially,biclustering algorithm can cluster genes and conditions at the same time,and mine the similar expression patterns of some genes under some conditions.Existing researchs have confirmed that biclustering is a typical NP-hard problem.Fortunately,intelligent optimization algorithm could be chose to solve this problem.Therefore,this thesis proposes a biclustering method based on intelligent optimization algorithm.The main work of the thesis includes:(1)Based on the improved bat algorithm(Improved Bat Algorithm Biclustering,IBAB),a biclustering method is proposed.Considering that the original bat algorithm is only suitable for solving continuous problems,this method introduces a V-shaped function to map continuous variables to discrete variables of 0 and 1,so that the bat algorithm can solve the biclustering problem with the 01 coding method.In order to improve the global search ability of the bat algorithm,this method introduces a dynamic decreasing inertia weight,which is used to improve the global search speed of the algorithm in the early stage,and reduce the local search speed of the algorithm in the later stage,so as to balance the global search and local search speed of the algorithm.Since the local search formula of the original bat algorithm is no longer applicable to the binary bat algorithm,this method introduces a mutation operator for local searching.For the evaluation of biclustering quality,the evaluation metric used in this thesis is Mean Squared Residue(MSR).The goal of the biclustering problem is to find biclusters with a lower MSR value and a larger volume,so the biclustering problem can be regarded as a multi-objective problem.This thesis combines the MSR and volume calculation formulas with different weights into a single function for converting a multiobjective problem into a single-objective problem.This method is tested on the yeast dataset,and the experimental results are evaluated by the CI index.The results prove that the algorithm has good performance.(2)Based on bat algorithm and genetic algorithm(Bat Algorithm and Genetic Algorithm Biclustering,BAGAB),a hybrid biclustering method is proposed.The bat algorithm has excellent global search ability,but its local search ability has obvious shortcomings,while the genetic algorithm and the bat algorithm are just the opposite.Therefore,this method combines these two algorithms and proposes a hybrid biclustering method.The effectiveness of this method is demonstrated by conducting relevant experiments on three datasets and comparing the three dimensions of MSR,ACV and volume with the results of the three classic biclustering algorithms CC,FLOC and DBF.