Prediction of Cancer Drug Sensitivity Based on GBDT-RF Algorithm

Yating Li, Jin Guo, Zongwen Fan

Abstract: Accurately predicting tumor drug sensitivity is important in drug development and selection. To address this issue, we propose a novel machine learning model, called GBDT-RF, using gradient boosting decision tree (GBDT) algorithm and random forest (RF) algorithm based on the drug sensitivity IC50 correlation data from the GDSC database. Through the prediction analysis of eight drugs, compared with the GBDT, RF, logical regression (LR), and support vector machine (SVM), our GBDT-RF algorithm has the best performance for predicting cancer drug sensitivity in terms of all metrics used. This shows that the GBDT-RF algorithm has certain advantages over the conventional machine learning models. Our proposed model can also provide some reference for medical decision-makers to predict tumor drug sensitivity.