Identification of Combination of SNPs Associated with Crohn's Disease Using a Balanced Univariate Marginal Distribution Algorithm

Detecting diseases associated SNPs is the central goal of genetics and molecular biology. However, highthroughput techniques often provide long lists of disease SNPs candidates, and the identification of disease SNPs among the candidates set remains timeconsuming and expensive. In addition, contrasting to the number of SNPs involved, the available datasets (samples) generally have fairly small sample size and the datasets used in these studies are often imbalanced (the number of cases and controls are not equal). Therefore, it is necessary to develop a robust tool to identify the disease associated SNPs. In this paper, we proposed a modified univariate marginal distribution algorithm (MUMDA) taking into account the imbalanced ratios of case/control for selecting the associated SNPs. In addition, the sparse vector encoding method was used in this paper. We illustrated the effectiveness of our algorithm using Crohn's disease (CD) dataset which includes 144 cases and 243 controls (each one has 103 SNPs). Our observations suggested that the proposed method could capture important features of the genetic architecture of CD and our algorithm outperformed other current methods.