Study On The Response Of Drugs Based On Generative Adversarial Networks

As the structure becomes more and more complex and the number of drugs is increasing,the research on drugs has entered a whole new stage.If we still use ordinary biological experiments to conduct research,it will undoubtedly be expensive and time-consuming.Therefore,it is imperative to conduct research on drugs by calculation methods.The research on the prediction methods of the effect of drugs is an efficient and novel drug research method.Although there are a large number of prediction algorithms for predicting drug effects,including treatment effects and related functional pathway,these algorithms introduce a large amount of prior knowledge while predicting drug effects,such as the drug similarity network and the network itself.Prejudice affects the final result.Therefore,there is a need for an unbiased method to predict the efficacy of drugs,and to obtain more fair results while being efficient.This article uses more objective gene expression profiles after drug action to measure the performance of drugs.At the same time,a prediction model of drug effect based on generative adversarial network is proposed in this paper.(predict-profile-conditional Generative Adversarial Nets(ppc-GAN)).The method in this paper first trains the selfencoder to compress the gene expression profile before and after medication;then integrates the encoder of the trained self-encoder into the model that generates the adversarial network,so that the amount of model parameters is reduced.The generators of the adversarial network are combined into a predictor.With regard to the prediction of gene expression profiles after drug action,the model was validated by experimental data before and after drug administration.The final results show that the predicted gene expression profile and the real gene expression profile have a strong linear relationship with a low error score.For specific examples,the predicted gene expression profile was used in this paper to verify gene regulation and treatment effects,and it has also performed well.At the same time,by deleting different parts of the model,it is observed that the design of each part of the model is meaningful,and it is helpful to improve the accuracy of the model and reduce the amount of parameters.The ppc-GAN algorithm not only predicts the gene expression profile after medication by using the gene expression profile before medication.When studying specific genes,the output of the model generator can be changed to the gene to be studied,and the model is retrained,which helps For improved accuracy.The compression part of the encoder can also be used in other models for data compression of gene expression profiles.This article proposes that the gene expression profile after medication is a new perspective to explain the effects of drugs,and also a brand-new perspective to explain the process of drug experiments through mathematical models.It is a very meaningful method for biological data generation and drug screening.