Classification Of Tumor Gene Expression Profiles Based On Sparse Theory

With the advent of DNA microarray technology, gene expression data of creature have explosive growth, attracting a large number of researchers into the field of bioinformatics,making bioinformatics received wide attention and research focus. Tumor samples classification method based on DNA microarray provides a new idea for the rapid and accurate diagnosis of cancer types. At the same time due to the small sample size and high noise characteristics of DNA microarrays, making this diagnostic method is facing major challenges.Sparse representation is popular in recent years of new technology, and by the attention of many researchers. Which has been widely used in the field of pattern recognition, such as face recognition, tumor classification, clustering, dimensionality reduction and so on.In this thesis, we focus on sparse representation techniques to study the classification problem on tumor gene expression data, and focus on the data of the high noise characteristics,proposed an effective noise reduction or de-noising algorithm elaborated algorithm process, and is compared with the conventional algorithm. Research work of this thesis are summarized as follows:Firstly, the use of robust sparse representation classifier for classification. The algorithm is an improved sparse representation classifier, which adding weight matrix and iterative algorithm to reduce weight value of noise and outlier in the data, that is to reduce the impact of noise or outlier of classification in order to achieve the purpose of noise reduction, and ultimately to improve tumor classification accuracy. Through experiments comparison with other algorithms, our method is feasible and effective.Secondly, the use of low-rank representation latent data preprocessing algorithm. The algorithm can remove noise and extract latent from data, while ultimately to improve the sparse representation classifier accuracy. Relatively large number of experimental results show that the algorithm is effective.Finally, this thesis concludes that the current classification of sparse representation exists some problems and future research needs to be done.