Prediction Of Anti-cancer Drug Sensitivity Based On Matrix Completion And Regression Model

Precision medicine is an inevitable trend in the treatment of cancer.How to extract key information from huge data,predict the therapeutic effects or side effects of anticancer drugs based on genetic information and other characteristics,and provide patients with the most appropriate treatment plan in time is the significance of precision medical treatment.Focusing on the above issues,this paper builds matrix completion,ridge regression,and weighted prediction models based on the two classic databases of Genomics of Drug Sensitivity in Cancer(GDSC)and The Cancer Cell Line Encyclopedia(CCLE)to effectively predict the sensitivity of anticancer drugs.This provides a theoretical basis for the screening of anticancer drugs.Firstly,considering the relationship between drug sensitivity data,the problem of anticancer drug sensitivity prediction is transformed into matrix completion problem,and a matrix completon prediction model is established.Using OptSpace algorithm and ten-fold cross validation method,the optimal parameters of the model are determined by calculating the Pearson correlation coefficient between the predicted value and the observed value.The matrix completion model has achieved good prediction results,and the results are higher than the popular ”cell line similarity network model”,”drug similarity network model” and”dual-layer integrated cell line-drug network model(ie cell line similarity network model and drug similarity network model weighted combination)”.Based on the hypothesis that there is a linear relationship between gene expression profiles and drug sensitivity data,the paper next uses the gene expression profile data as input features to establish a single drug ridge regression model with a 10-fold cross validation and Pearson correlation coefficients to screen out The marker genes(233?12535)that are strongly associated with drug sensitivity greatly reduce the data dimension.Although the model prediction results are lower than the”drug similarity network model” and ”dual-layer integrated cell line-drug network model”,it is superior to the ”cell line similarity network model.” In addition,the gene with the largest1000 regression coefficients for each drug in the ridge regression model was used to perform on-line analysis of David's gene function.Finally,a combing matrix completion-ridge regression weighted model is built based on matrix completion and ridge regression results to achieve the complementary advantages of both,and the prediction effect is particularly prominent.For at least 83% of drugs,the Pearson correlation coefficient between the predicted and observed values is higher than the ”dual-layer integrated cell line-drug network model”.It can be seen that the combing matrix completion-ridge regression weighting model can be used as one of the alternative tools for anticancer drug sensitivity prediction.