Fig. 1
From: Predicting drug response of tumors from integrated genomic profiles by deep neural networks
Illustration of DeepDR. (a) Model overview. Mutation and expression data of TCGA (n = 9059) were used to pre-train two autoencoders (highlighted in blue and green) to extract data representations. Encoders of the autoencoders, namely mutation encoder Menc and expression encoder Eenc, were linked to a prediction network (P; denoted in orange) and the entire network (i.e., Menc, Eenc, and P) was trained using CCLE data (n = 622, of which 80, 10, and 10% used as training, validation, and testing, respectively) to predict the response to 265 drugs. (b) Architecture of the neural networks. Numbers denote the number of neurons at each layer