Fig. 3
From: Deep convolutional neural networks for pan-specific peptide-MHC class I binding prediction
The DCNN architecture. The DCNN architecture is closely based on of the popular DCNN architecture proposed by Simonyan and Zisserman. Three convolution blocks with two convolution layers and a max pooling layer are concatenated, and three classification layers are then connected to the ends of the network. The dropout was next applied to each convolution block as a regularization. ReLU was used for the nonlinear transformation of the output value of each convolution layer