Research On Significant Genes Selection Method Based On PSO Algorithm

Significant genes selection plays an important role in tumor recognition. The feature gene selection could not only avoid the curse of dimensionality, but also identify the most significant genes. Because of its relative superiority to the traditional optimization algorithm, Particle Swarm Optimization (PSO) algorithm has recently been more widely applied to the feature gene seletction problem.This paper proposes two methods of significant genes selection base on PSO algorithm. Both approaches are according to this framework:first using a filter method to remove the genes that are irrelevant to tumor recognition and then adopting PSO algorithm to remove redundant genes.The support vector machine(SVM) is used as an evaluator to estimate performance of the selected significant genes for tumor diagnosing.The first feature gene selection method is based on the immune vaccine genetic PSO algorithm. Firstly, introducing genetic PSO algorithm into the field of gene selection. Secondly, in order to fix the blind search provided by the crossover and mutation operator in the genetic PSO algorithm, introducing vaccine mechanism to inhibit degeneracy during evolution. Experiment results show that 100% and 96.77% of 10-fold cross-validation accuracy has been achieved by only five and eleven genes for leukemia, colon tumor datasetsThe second feature gene selection method is based on the binary Quantum behaved Particle Swarm Optimization (BQPSO) algorithm. Firstly, introducing Quantum behaved Particle Swarm Optimization (QPSO) algorithm into the field of gene selection. But the QPSO can only solve the continuous optimization problem,so we redefine the particle position updating formula. For leukemia, colon tumor datasets, experiment results show that 100% and 96.77% of 10-fold cross-validation accuracy has been achieved by only five and seven genes.