Prediction Of The Synergy Of Anti-cancer Drug Combinations Based On High-throughput Sequencing And Deep Neural Network

For specific cancer types,finding a synergistic drug combination is essential to improve cancer efficacy.In this paper,based on the pharmacogenomic big data and high-throughput drug combination synergy score,a computational model is constructed to predict the synergy of anticancer drug combinations(drug-drug combination).The research has important guiding significance for synergistic drug screening experiments,and helps to shorten the clinical research process of experimentally determined drug combinations.Based on the gene expression information of cancer cell lines and the three types of characteristics of drugs,this paper combines the filtering with the wrappering method to extract the feature genes of genomics data,and proposes a prediction model based on variance ranking and deep neural network(VarDNN-EXP).Used five-fold nested cross-validation and leave-one-out method to predict the synergy of four types of "cell line-drug combination",including "old cell line-new drug combination","new cell line-old drug combination",“old cell line-new and old drug combination" and "new cell line-new and old drug combination".The results show that the prediction performance of the VarDNN-EXP model is not only better than the existing classic DNN model,but also the prediction results of the "new cell line-new and old drug combination discussed for the first time are consistent with the existing conclusions.In addition,VarDNN-EXP can identify biomarkers closely related to the occurrence and development of cancer,and enhance the biointerpretability of the model.In order to further improve the predictive effect of the model,this paper adds gene mutation data and copy number variation data,and constructs VarDNN combined model(VarDNN-COMB)based on back-end combination method.That is,for gene expression,gene mutation,and copy number variation data,the VarDNN calculation model is constructed separately,and the synergy scores of the three types of predictions are weighted and combined to obtain the final prediction result.The results show that in the synergistic prediction of“old cell line-new drug combination","new cell line-old drug combination",and "old cel1 line-new and old drug combination",the prediction performance of VarDNN-COMB is significantly better than the VarDNN model of single-class data and better than the classic deep neural network(DNN)model.Through enrichment analysis,it is found that the feature genes identified by VarDNN-COMB are closely related to the occurrence and development of cancer,providing a reference for subsequent studies on the prediction of anti-cancer drug combinations synergy.