A Study Of Gene Selection Method Based On Generative Adversarial Networks And Swarm Intelligence Optimization

Cancer has become one of the major diseases threatening human life and health.Gene selection of microarray data can effectively analyze and identify pathogenic genes and provide important references for the prevention and treatment of cancer.Gene expression profile is a kind of data set with high dimension and small samples.The gene selection method based on swarm intelligence algorithm can relieve the pressure caused by high dimension to some extent by its strong search ability.However,the lack of samples still leads to the poor effect of most gene selection methods.At present,the research methods are mainly focused on the improvement of dimensionality reduction algorithm,and the research on sample size is not concerned enough.Due to the above problems,this paper uses the method based on the combination of generative adversarial network(GAN)and swarm intelligence algorithm(SI)to conduct gene selection.This method improves the sample size of microarray data by generating samples,and obtains better gene subsets through the powerful search performance of swarm intelligence algorithm,which provides a new idea for the research of gene expression profile data.The main work of this paper is as follows:(1)In view of the defects of traditional gene selection methods,this paper proposes a gene selection method(y-CAGN-BPSO-ELM)based on the generative adversarial network and binary particle swarm optimization(BPSO)algorithm.Firstly,according to the characteristics of microarray data,this method uses the CGAN model for sample generation,and at the same time to add constraints on condition variable y into the discriminator model,so as to improve the sensitivity of the generated model to different types of labels.Finally,the added sampled data is selected by BPSO algorithm,and uses the classification results of extremely learning machine(ELM)to guide the selection process.This method can effectively generate expression profile samples and select gene subsets with high classification accuracy.(2)Aiming at the problems of mode collapse,slow convergence and easy convergence to local optimum in face of multi-classification problem,A gene selection method based on double discriminator for GAN and hybrid swarm intelligent optimization(DDCGAN-ABPSO-ELM)is proposed.Based on the GAN network model,this method proposes a generative model with a double discriminator(Double Discriminator),which uses two different discriminators to set the weights on the authenticity and diversity of the generated samples to obtain a higher degree of authenticity and more uniformly distributed gene expression profile samples.Then,through the population sharing mechanism to expand the search range and improve the convergence accuracy with the combination of PSO algorithm and ABC algorithm.The experimental results on multiple binary classification and one multi-class gene expression profile data prove that this method can simultaneously solve the problem of high gene expression profile dimension and small sample size,generate samples with higher authenticity and diversity,and choose a subset of genes with higher classification performance and stronger correlation.(3)Based on the algorithm proposed in this paper,a set of gene selection system is designed and implemented.The feature selection function of different gene expression profile data sets is realized by different gene selection methods,and the optimal gene subset is displayed and its classification ability is verified.