Research On Gene-Gene Interaction Identification Method Based On Support Vector Machine

Genome-wide association study usually use single nucleotide polymorphisms as a marker for complex disease analysis.However due to the genetic data with a small sample size,high dimension,large noise characteristics,the way of traditional experiment research of the gene-gene interaction is time-consuming,laborious,and costly.With the help of the data mining technology,analysis gene-gene interaction accuratelyfor complex disease.And it has important significance for exploring or finding the etiology of susceptibility genes.In this study,we propose a approaches/algorithms based on support vector machine called SVMITER.The main contents of this dissertation are described as follows:(1)SVMITER algoritm.First we proposes a new algorithm SVMITER.It based on support vector machine andcartesian product.Algorithm using support vector machine to early screening SNP,after screening SNP using the cartesian product algorithm for SNP combination.Then we use simulated data comparison of the current main three gene-gene interaction analysis method:BOOST,MDR and RF.SVMITER on classification performance and computing time are better.(2)Low order gene-gene interaction research.We use SVMITER method for detecting low-order gene-gene interaction.Our research using simulated data and real data contrast experiment performance.Comparing with the existing methods and two case studies show that SVMITER algorithms identify performance is better than the BOOST in the simulated data.Furthermore,two case studies demonstrated that SVMITER can be applied rapidly to accurately identify gene-gene interaction.It can accurately identify the SNP combination rs380390 and rs 1329428,etc.A ten cross-validation and independent tests using datasets demonstrated that SVMITER outperformed well.(3)High order gene-gene interaction research.We use SVMITER method for detecting high-order gene-gene interaction.Comparing with the existing methods and onecase studies show that SVMITER algorithms identify performance is better than the BOOST in the simulated data.Furthermore,one case studies demonstrated that SVMITER algorithm in high order can identify 5 order SNP combinations,and recognition performance is still better than BOOST algorithm.