Feature Selection Based On Frequency Of Mutual Information And Its Application To SNP Association Study

SNP (Single-Nucleotide Polymorphism) is a newly kind of genetic marker. It is a common form of genetic variation and it has been detected that SNP almost reached 90% of all genetic variation. The research of SNP has significance for post-genome era. However, because of its large presence in the human genome, and the existence of a large number of redundant and irrelevant SNP that is unrelated to disease, we find disease-related SNP difficult. These problems are usually resolved by feature selection. Feature selection has received a significant development since 1970's. There are many feature selection method. This paper is trying to find out these disease-related SNP by feature selection.In this paper, we use the mutual information as a feature selection of the evaluation function, and propose a frequency-based heuristic method to search the features of the candidate set (Branch Matrix Search algorithm). Later, we made two sets of experiments. One set of experiments is to compare algorithm of this paper with two others algorithms (ME method and mRMR method). In another set of experiments we discussed two influences:influence of SNP of different models and influence of inter-SNP of one model.