Prediction of Cancer Drug Effectiveness Based on Multi-Fusion Deep Learning Model
Qian Li, Jie Wen Huang, Hongming Zhu, Qin Liu · 2020 10th Annual Computing and Communication Workshop and Conference (CCWC) · 2020
Oncogenomics is the premise of precision medicine, which has attracted the attention and research of many scholars in recent years. Among them, predicting the response of cell lines to drugs is a very important topic. Because understanding the response of cell lines to drugs can not only save a lot of time in drug screening, but also promote the reuse of existing drugs that have been approved by the Food and Drug Administration(FDA) and other regulatory agencies. Herein, this paper proposed a new multi-fusion neural network model. In this model, we first use the Convolutional Neural Network(CNN) to capture the gene expression features of the cell line and the molecular descriptor features of the drug, and then use the resulting abstract features as the input data of the long and short-term memory neural network(LSTM) for drug response prediction. By comparison with some traditional machine learning algorithms and CNN model, we found that our model can improve the prediction accuracy. In addition, compared with previous works on modeling a single drug or a single cancer type, in the design of this model, we extended the application categories to human tissues, where a tissue consists of multiple TCGA types. This paper provides a new method for drug response prediction and provides some guidance for the screening of effective anti-cancer drug.