Computational Models Of Anticancer Drug Response Prediction Based On High-throughput Sequencing

The lack of anti-cancer drug sensitivity data will have an important impact on the subsequent analysis of cancer data.High-throughput sequencing technology provides the possibility to build a computational model and effectively predict the response of anticancer drugs.Based on two classical data sets,Cancer Cell Line Encyclopedia(CCLE)and Cancer Drug Sensitivity Genomics Database(GDSC),this paper constructs two types of computational models to predict anticancer drug response,including the prediction of the sensitivity of anticancer drugs and two classification prediction of sensitivity and resistance of anticancer drugs,in order to provide a new research perspective for the prediction of the response of anticancer drugs!Based on the existing rationality assumption: similar cell lines have similar sensitivities to target drugs,and similar drugs have similar sensitivities to target cell lines.This paper comprehensively considers the gene expression and gene mutation characteristics of cell lines,and gives a new definition of cell line similarity;Using the chemical structure of drugs,combined with the drug similarity measurement method,a prediction model of anticancer drug sensitivity(C-D KNN)based on cell line-drug KNN is proposed,and the prediction results obtained are significantly better than some existing classic models.In order to further improve the predictive effect of the computational model,this paper equates the problem of “predicting missing values of anticancer drug sensitivity ” to the problem of “matrix completion” and proposes an anticancer drug response prediction model(NNRM)based on kernel norm regularization.The augmented lagrangian function is used to transform the constrained optimization problem into an unconstrained problem,and the singular value decomposition soft threshold estimation(Soft-impute)and alternating direction multiplier method(ADMM)are used to solve the problem,which not only reduces the dimension of the optimization problem,but also speed up optimization.The sensitivity prediction result of NNRM is obviously better than existing models.In order to further verify the prediction performance of NNRM,the sensitivity observation values of target drugs to all cell lines were sorted,and the “sensitivity-resistance dichotomy” prediction problem is constructed by setting threshold values.The classification results based on NNRM sensitivity prediction values are obviously superior to the popular CDCN model and SRMF model.In particular,it is worth mentioning that NNRM does not need to consider cell lines and drug characteristics,only from the observation data of drug sensitivity,it can realize the prediction of cross cancer,large-scale and batch anti-cancer drug response,greatly reducing the data requirements of the calculation model.