Fig 2. Structure of a convolutional neural network in DeepDrug3D.

The network consists of (A) an input voxel followed by (B) two convolutional layers with leaky ReLu activation functions, and (C) a series of dropout, pooling, fully connected and softmax layers.