Feature Selection for SNP Data Based on SVM-RFE and AGA

Within the association analysis of high-throughput SNP data, it has been faced two main challenges, the super-high dimension data with less samples and the complex interaction between SNPs of genetic disease. This paper proposes a feature selection method, which combines Support Vector Machine based Recursive Feature Reduction method(SVM-RFE) and Adaptive Genetic Algorithm(AGA). Under the premise of retainning the correlation between SNPs, it significantly reduced the optimization space of critical SNPs using SVM-RFE, and then quickly found the SNPs which distinguish sample type most effectively using AGA. Compared with the Chi-square analysis and modified Relief algorithm, experimental resualt shows that the suspect SNPs selected by this method has a better ability to distinguish the type of samples. This method provides a feasible way for SNPs association study, and screen out an appropriate scale of SNP set for further biomedical research.