Research On Multi-Objective Evolutionary Computation For Biclustering In Microarray Gene Expression Data

Microarray techniques can measure the expression levels of thousands of genes simultaneously. Those microarray data provide massive amount of information, which is leading to the development of sophisticated algorithms capable of extracting novel and useful patterns from a biomedical point of view. Biclustering approach is a very useful technique of data mining from microarray data, and show strong advantage in many applications. While mining biclustering from microarray data, many objectives conflicting with each other need be optimized simultaneously, which Multi-Objective optimization (MOO) is being the very good approach of solving biclustering.Evolutionary Computation (EC) is a field of research dedicated to the study of algorithms which maintain some form of solution memory which is used to bias future solution creation. Some common examples of algorithms fitting into this broad definition of EC contain Genetic Algorithms (GA), Genetic Programming (GP), Evolutionary Strategies (ES), Differential Evolution (DE), Simulated Annealing (SA), Particle Swarm Optimization (PSO), Artificial Immune Optimization (AIO), and Ant Colony Optimization (ACO). EC, as a general algorithm methodology, mainly focuses on the diversity of solutions which is very important to MOO.In the recent thirty years, it is a common data mining approach to simulating natural phenomena such as evolutionary, genetic and immune. Evolutionary algorithms(EA) can find global optimal solutions from microarray data. As optimizing many conflicting objectives (in the case of biclustering, such as the size and of the homologous of cluster), multi-objective evolutionary computation algorithms (MOECA) is proposed to find globally efficient clusters of microarray data.This paper mainly aims on the research on MOECA to solving biclustering o f microarray data, such as multi-objective evolutionary computation biclustering algorithms, multi-objective particle swarm optimization biclustering, artificial immune optimization biclustering and mulli-objective ant colony optimization biclustering.First, this paper provides a survey and the application area of biclustering. The author analyses the challenge of microarray data biclustering. The current MOO algorithms and the application in bioinformatics are provided. Then the premise of MOO biclustering algorithm is described.This pa pers ana lyses t he cur rent evolutionary computation algorithms a nd multi-objective evolutionary bi clustering algorithms, and summarizes the algorithm framework. Introducing a local search strategy, the author proposes Multi-Objective Evolutionary T riClustering (MOE-TC) to mine 3D c lusters f rom G ST (Gene-Sample-Time) data. Then usingσselecting and (?)-dominance strategy to quicken t he convergence of the algorithm, the author proposesσselection based Multi-Objective Evolutionary TriClustering algorithm. Experimental analysis is provided in the last of the section.Particle Swarm Optimization (PSO) is a heuristic search technique that simulates the movements of a flock of birds which aim to find food. The rapid convergence and relative simplicity of PSO and the fact that is a population-based technique have made it a natural candidate to be extended for MOO. This paper adapts multi-objective particle swarm optimization to find the global optimal solution of biclustering. Combining (?)-dominance with local search technique, this paper proposes Multi-objective Particle Swarm Optimization Biclustering (MOPSOB) to mine maximal significant biclusters with lower mean squared residue. To further improve the diversity of optimal solutions, applying crowding distance based update strategy, the author proposes Crowding based Multi-objective Particle Swarm Optimization Biclustering, CMOPSOB), which has better performance than MOEAB algorithms.Recently, while solving MOO, the use of artificial immune system can improve search ability and adaptability, and quickens the convergence of the algorithm and enhances the diversity. After providing the survey of the current artificial immune algorithms and multi-objective immune optimization approach, based on the immune response principle of artificial immune system, this paper extends dominance relation and performs the mechanism of crowding computation and proposes a novel multi-objective immune biclustering (MOIB) algorithm to obtain many Pareto optimal solutions distributed onto the Pareto front. Experimental results on real datasets show that our approach can effectively find more significant biclusters than other biclustering algorithms. Ant Colony Optimization algorithms(ACO) simulate the biology behavior of ant colony finding food, and have been shown to be effective problem solving strategies for a wide range of problem domains, including MOO. Multi-Objective Ant Colony Optimization (MOACO) mainly focuses on solving the multi-objective combinatorial optimization problems, and the biclustering problem is combinatorial. This paper, therefore, incorporates local search technique into ACO, and proposes a novel Multi-Objective Ant Colony Optimization Biclustering (MOACOB) algorithm to mine biclusters from microarray dataset. Then incorporates crowding update technology into MOACOB, this paper proposes a novel Crowding based MOACO Biclustering (CMOACOB) algorithm. Experimental results are shown on two real gene expression dataset.In summary, this thesis studies deeply multi-objective evolutionary computation biclustering of microarray data, and develops several effective biclustering algorithms for microarray gene expression data, which has academic and practical value for advancing the theory and practicability of clustering in high dimensional data.