Hybrid System Design For Microarray And Mass Spectrometry Data Analysis

With the development of the high-throughput technologies, microarray and mass spectrometry technologies are now taking place of the traditional methods and transferring the way researches are done. One important application of such technologies is in human diseases studies in which microarray and mass spectrometry are often utilized to comparing and measuring the difference between diseases or certain biological traits and normal samples. Despite its great promise, such kind of technologies challenged our data analysis ability. The challenge originates from the nature that the datasets generated by such high-throughput experiments are often with large number of features while the number of samples is limited. Therefore, how to select a feature subset which can yield high sample classification accuracy is of great importance. After nearly a decade of research, several methods and algorithms have been employed and developed to tackle these problems. Yet, large number of experimental results suggested that no method can consistently perform superior on any datasets. To different datasets and different experiment strategies, different analysis methods may perform unevenly. In order to compensate the drawbacks of each methods while integrate the strengths of them, the development of hybrid system has been identified as a promising way.In this thesis, we discuss how to develop the hybrid system for microarray and mass spectrometry datasets analysis. Firstly, we propose a multi-objective genetic algorithm (MOGA) based ensemble classifier system for the analysis of micoarray datasets. The experiment results of both binary-class datasets and multi-class datasets generated from various micoarray studies demonstrated the usefulness of this hybrid method. Secondly, the above hybrid system is further integrated with the k-mean clustering algorithm, forming a clustering based feature selection hybrid system. Based on the central dogma of biology, we successfully adopted the idea in microarray data analysis into the analysis of mass spectrometry data. The experiment of two benchmark mass spectrometry datasets illustrated the effectiveness of this clustering based feature selection hybrid system.