Anticancer Drug Response Classification Based On Deep Neural Network And Support Vector Machine

Different patients may have different responses to the same anti-cancer drugs.To understand the differences in the responses of patients about anti-cancer drugs is of great reference value for cancer precision medicine.The continuous improvement of high-throughput sequencing data provides a good data foundation for constructing anti-cancer drug response classification prediction models,and then mining the hidden information behind the data.Based on two classic data sets,Cancer Cell Line Encyclopedia(CCLE)and Genomics of Drug Sensitivity in Cancer(GDSC),this paper constructs two classification models for predicting the sensitivity or inhibition of cancer cell lines to anti-cancer drugs,which provide alternative tools for anti-cancer drugs response prediction and biomarker identification.Based on the max-relevance and min-redundancy algorithm(mRMR)and deep neural network(DNN),this paper constructs the mRMR-DNN model.First apply variance sorting to eliminate a large number of redundant genes,and then extract 500 characteristic genes with the help of mRMR algorithm,and then use the gene expression information of characteristic genes to construct DNN with dynamic regulation algorithm(Adagrad),and predict cancer cell lines by five-fold cross-validation method classification of responses to specific drugs.The experimental results show that the prediction results of mRMR-DNN on the two classic data sets are better than the deep response forest model.In order to further evaluate the prediction performance of the model,mRMR-DNN was applied to the three representative drugs with more missing sensitivity values in GDSC,and the cell lines with missing response labels were classified and predicted.Draw a conclusion :The three representative drugs are more sensitive in mutant cell lines,and are completely consistent with the reaction classification observed in the experiment,which once again illustrates the predictive effectiveness of mRMR-DNN.Although mRMR-DNN has high prediction performance,its computational complexity is relatively large.In order to solve the problems above,this paper proposes a mRMRSVM model based on mRMR and support vector machine(SVM),using the same feature extraction method and using SVM to predict anti-cancer drug response classification.The experimental results show that the prediction results of mRMR-SVM on the two data sets are better than mRMR-DNN.In addition,applying mRMR-SVM to three types of tissues(hematopoietic and lymphoid tissues,lung tissues and skin tissues)containing more cell lines in CCLE,the prediction results are significantly better than the existing SVM model.Finally,the literature search verifies that many of the feature genes extracted in this paper are closely related to the occurrence and development of cancer,and once again that the feature genes extracted in this paper can be used as an effective factor for predicting the response of anti-cancer drugs.