Gene Selection In Microarray Database Based On Graph Energy

DNA microarray technology can quickly detect tens of thousands of genes, and it has given an overall understanding at the genetic level. In the last few years, more and more scholars apply microarray to tumor and cancer research. DNA microarray has a very high dimensionality (large number of genes) with a small number of samples, and it is important to select a few genes which are strongly related with cancer for finding tumor pathogenesis and gaining significant insight into the mechanism of disease. The author devotes herself to getting a higher accuracy rate of classification with few typical genes. The main contributions of this research are summarized as below:The energy levels of the molecular, describing how the π-electrons move within the molecule, coincides with the eigenvectors of the graph. The graph energy is an important variable describing molecular properties. In this paper, the author introduces the applicability of graph energy to the gene selection for the first time, and proposes a new formula named SNRGE utilizing graph energy combined with modified single-to-noise-ratio. In the proposed gene selection process, k-means clustering method is applied to group entire genes into k categories first, and the SNRGE formula is employed to extract the informative genes from each category. This method is used to select a small subset of genes from broad patterns of gene expression data, recorded on DNA microarray data, and the effectiveness of SNRGES is measured by the SVM classification rate. The author demonstrates particularly in the database of colon that it can obtain excellent effect when drawing the graph energy into the selection of gene.To validate the performance of the proposed method, the author will make a systematic and comprehensive analysis. The author performs several cross tests with the traditional signal-to-noise ratio (SNR) method to compare gene selection methods and classifiers. Then she uses three other SNR formulas proposed in other papers to replace SNRGE in this method to testify the effectiveness of SNRGE. Next, different classifiers are employed to validate the classification rate of various gene selection techniques. And in the fourth part, the author checks the relevance to colon cancer diagnosis with the selected genes. And then the method is applied to four different DNA microarray databases. Though the five tests above, it indicates that SNRGES is capable of achieving better performance than previous studies.