Deep Graph and Sequence Representation Learning for Drug Response Prediction

Xiangfeng Yan, Yong Liu, Wei Zhang

Abstract: Drug response prediction plays a crucial role in personalized medicine and drug discovery. Many deep neural networks have been proposed for better drug response prediction. However, these methods only represent drugs as strings or represent drugs as molecular graphs, failing to capture comprehensive information about drugs. To address this challenge, we propose a joint graph and sequence representation learning model for drug response prediction, called DGSDRP. We use convolutional neural networks (CNN) to obtain local chemical context information from the drug sequences and a fusion module based on CNN and Bi-LSTM to capture the features of cell lines. Furthermore, we use graph convolutional networks (GCN) to extract topological structure information from the molecular graphs. Finally, we concatenate all representations through several dense layers and end with a regression layer to predict the response value. Extensive experimental results show that our model outperforms the current state-of-the-art models in terms of RMSE and \(CC_p\).