A computational model for anti-cancer drug sensitivity prediction

Various methods have been developed to build models for predicting drug response in cancer treatment based on patient data through machine learning algorithms. Drug prediction models can offer better patient data classification, optimising sensitivity identification in cancer therapy for suitable drugs. In this paper, a computational model based on Deep Neural Networks has been designed for prediction of anti-cancer drug response based on genetic expression data using publicly available drug profiling datasets from Cancer Cell Line Encyclopedia (CCLE). The model consists of several parts, including continuous drug response prediction, discretization and a drug sensitivity result output. Regularization and compression of neuron connections is also implemented to make the model compact and efficient, outperforming other widely used algorithms, such as elastic net (EN), random forest (RF), support vector regression (SVR) and simple artificial neural network (ANN) in sensitivity analysis and predictive accuracy.