Research On Drug Sensitivity Prediction Method Based On Iterative SIRS

Cancer patients respond variously to each cancer therapy because of genetic factors,environmental causes,and disease diversity.Predicting of drug response based on genetic measurements is a crucial task in the research of personalized cancer therapy that has attracted increasing recent attention from various domains such as machine learning,data mining and computational biology.By predicting the drug response to cancer,oncologists obtain full understanding of the efficient treatments on each patient,which leads to better personalized cancer therapy.Existing approaches to predicting drug sensitivity rely mainly on profiling of cancer cell line panels that have been treated with different drugs and selecting genomic or functional genomic features to regress or classify the drug response.However,it is still challenging to precisely predict the effectiveness of therapeutics in humans within a complex genomic and molecular context.In this work,we proposed an iterative sure independent ranking and screening(ISIRS)scheme to select drug response-associated features and applied it to the Cancer Cell Line Encyclopedia(CCLE)dataset.For each drug in CCLE,we considered different genetic information including copy number alterations,mutation and gene expression.We selected up to 50 features based on a marginal measure.The importance of genomic features was assessed based on a marginal measure which concerns with the conditional distribution of the response given the features.We fit a liner regression model based on the selected features by ISIRS and predict drug response by the estimates of ordinary least squares.And we can detect some features that their cok rankings are very low,but are significant for the regression model according to the t-test of regression.Sparse regression modeling based on the selected features showed that our prediction accuracies are higher than those by iterative sure independence screening for most drugs.We find that all the MSEs of ISIRS are lower than those by ISIS with p-value 5.96e-08 by Wilcoxon-test and proportional reductions in MSE of some drugs are much bigger such as Paclitaxel?Panobinostat?Irinotecan.Thus,person correlation coefficients(PCC)of real and predicted sensitivities by ISIRS are higher for most drugs in CCLE dataset.Moreover the increasement of PCCs by ISIRS compared to ISIS for three drugs(L.685458,Nilotinib and Paclitaxel)was greater than 0.05.And the PCC for TKI258 is improved from 0.42 to 0.46.