Microarray Data Analysis Based On Locality Sensitive Criterion With Sparsity

With the progress of The Times, the core of the biological research is changingfrom gene sequence to gene function. The cDNA microarray technology makesit possible to simultaneously measure the expression level of genes. This tech-nology spurs the research on biomarker identification and cancer classificationusing gene expression data. However, how to e?ective exploit the useful infor-mation from these unprecedented amount of microarray data by computationalmethods is an open and challenging issue. In recent years, various discriminantanalysis and variable selection methods have been used for analysising geneexpression data. So more and more biomarker identifications which identifybase genes been critical to tumor are coming. This makes it easy to diagnoseand explain.Our method is based on locality sensitive semi-supervised feature selectionand sparseness and using this for biomarker identification and diagnosis of can-cer. After learning the following methods: Locality sensitive semi-supervisedfeature selection, Locality sensitive discriminant analysis and Sparse PCA, wepropose a novel feature selection–Microarray data analysis based on localitysensitive criterion with sparsity, which combine Locality sensitive methods andSparse methods. So, on the one hand, In order to obtain good e?ect of cancerclassification, we use the advantages of locality sensitive method, that is we can discover the local manifold structure apart from the globle structure for dis-criminant analysis, eventually the nearby points with the same label are closeto each other while the nearby points with di?erent labels are far apart. On theother hand, we inspect our problem in a regression-type view and then use theelastic net to produce sparse maximum feature, it is significant for biomarkeridentification that these feature can be interpreted intuitively.