Feature Selection for Single Nucleotide Polymorphisms using Parallal Genetic Algorithm

Abstract

Abstract. The application of statistical machine learning methods to study the association between large scale single nucleotide polymorphism (SNP) and complex diseases is facing the "Curse of dimensionality". The first task is to reduce the large scale SNP to smaller sets. For this reason, a multiple genetic algorithm is proposed for feature selection of single nucleotide polymorphism. For the first time, a new method is proposed to measure the degree of association between SNP and disease by mutual information, and as the fitness value of genetic algorithm (GA), the candidate feature SNP set is obtained by the use of genetic algorithm. In the SNP experiment on simulation data and the maximum entropy (ME) method for performance comparison shows that, this method can reduce the SNP collection and disease related SNP, while retaining the disease associated with SNP, SNP provides data for further research on the suitable scale, this method can be used in medium or large scale SNP set.