Predicting Drug Response Using Two Factors From Cell Lines-Drug Sensitivity and Basal Gene Expression

Patient drug response plays a critical impact in clinical pharmacology. To accurately predict clinical drug response, we propose a prediction method using two-factor cell lines-drug sensitivity and basal gene expression. We input the baseline gene expression data from a large panel of cell lines and use several machine learning models (ridge regression, random forest, SVM) to predict chemotherapeutic response in patients. Specifically, first, we fit 80% of the gene expression and drug sensitivity data into our models for training. Next, we evaluate the accuracy and effectiveness of our prediction method by applying the rest 20% data to our trained models, which yields a good drug sensitivity prediction. Last, we reduce the number of genes with two different types of feature selection, dramatically decreasing the training time without significant changes in prediction accuracy. Our experimental results demonstrate that our models with feature selection are good tools for drug response prediction.