Research On Key Technologies Of Generative Adversarial Networks In The Field Of Bioinformatics

Generative adversarial networks(GANs)are currently an important research area in deep learning.Mainly based on the idea of zero-sum game,GANs consists of a discriminator neural network and a generator neural network,and minimax algorithms is used for training.The purpose of GANs is to find the potential rule of the data sample in order to estimate the probability distribution of the data and generate new data samples based on the learned distribution.The GANs show great research value and have application prospects in the fields of computer vision,information security,and voice processing.In the field of bioinformatics,with the proliferation of biological data,precision medicine,the generation and prediction of gene sequences,and the synthesis of proteins all require researchers to make in-depth analysis of massive data.In the past,traditional methods such as wet-lab experiments and chemical analysis often had the disadvantages of high financial resources,large workload,and low efficiency.Facing this kind of problem,using data mining technology to discover data rules and obtain the essential characteristics of data are the most effective analysis methods.However,there still few study on the GANs in the field of bioinformatics.Therefore,this paper is based on the GANs algorithms to learn the rule of data existence.This article first elaborates the related technologies of deep learning,the principle and advantages of the GANs.It further analyzes the problem,such as the GANs do not converge in the results,the gradient disappears,uncontrolled results,and cannot trained discrete data in bioinformatics.Variational Auto-Encoder(VAE)can deal with discrete data,and its loss function can constrain the results produced by the model.Then,in this paper,a VAE and Generative adversarial networks combined fusion model,called DGAN-VAE(Double Generative adversarial networks with VAE),is proposed,where the VAE is introduced to optimize the GAN.A new neural network structure was designed for image and sequence data.Experiments have shown that DGAN-VAE can work well in pictures and gene sequences in the field of biological information.Our work will benefit the key technical issues and feasibility in the field of bioinformatics for the Generative adversarial networks.