Research Of Gene Interaction Detection Algorithm Based On Markov Blanket

Bayesian Network is a graphical network based on probabilistic reasoning, and a Markov blanket of a Bayesian network is a method to find the set of variables according to conditions of association between variables. Many studies have found that the Markov Blanket method in Bayesian network is suitable for epistasis detection in genome-wide association study. In recent years, a series of epistasis detection algorithms based on Markov blanket have been proposed, however, these algorithms are inefficient and their false positive rates are high for the large-scale data of GWAS. Faced with these problem, we intensively study the epistasis detection algorithm based on Markov blanket in this paper.In order to improve the performance of the epistasis detection algorithm based on Markov Blanket, we propose an optimized epistasis detection algorithm--OMBED(Optimized Markov Blanket for Epistasis Detection). This algorithm consists of three phases: Remove phase,Forward phase and Backward phase. In the Remove phase, independent variables are removed from the candidate set according to conditions of association between variables. In the Forward phase, using G2 test as the measure of independence between variables, we obtain the minimal Markov Blanket variable set by removing the weakly associated variablesfrom the candidate variable set and adding the strong associated variables to the Markov Blanket variable sets. In the Backward phase, the false positive variables are removed from the Markov Blanket variable set. In the Forward phase, we optimize the operations of adding and removing on the basis of the original algorithm, the number of G2 tests and the complexity of the algorithm are reduced. Experimental results on a series of simulated data sets and real data sets show that the algorithm is more efficient and reduces false positive rate.