Research On Epistasis Detection Methods In Genome-Wide Association Studies

With the completion of the international HapMap project, and the developmentof high-throughput technologies, human genetics is entering a new period. At thesame time, the development of genome-wide association studies unveiled new preludefor research on human trait/complex diseases. Epistasis is defined as the interactionbetween two or more genes (or single nucleotide polymorphisms). As an importantpart of complex trait genetic research system, epistasis detection has been widelyconcerned by researchers. A lot of epistasis test algorithms are proposed in recentyears. While for geneme-wide, large-scale data, there are still some problems, such aslow power and high false positives. In this paper, we focus on designing epistasisdetection algorithm in genome-wide case-control studies.Considering the lack of existing epistasis tests based on Markov blankettwo-step methods, we propose a novel epistasis detection algorithm, IMBED(Improved Markov Blanket for Epistasis Detection), and apply to genome-wideassociation studies. We use G2as the measure of independence between variables,define formulas of number of variables want to be removed. By removing variablesindependent and weak associated target variable, the search space is reduced.Experimental results show that the new method could detect causal SNPs moreefficient, less false positives, higher sample efficient, and can be used ingenome-wide case-control studies.In order to reduce the data requirements and improve the detection power, weproposed a new epistasis detection algorithm, PCED(Parents and Children basedMarkov blanket for Epistasis Detection), based on divide-and-conquer approach ofMarkov blanket. The problem of epistasis detection is divided into severalsub-problems, and all sub-problems are solved. We detect one disease associatedsingle nucleotide polymorphisms (SNP) each time, and detect k times to finish thedetection of k-way casual SNPs in genome-wide case-control studies. We evaluatedPCED algorithm on both simulated datasets and real dataset. Experimental resultsshow that the epistasis detection power has been further improved.