Fig. 2
From: Deep learning for cancer type classification and driver gene identification
The architecture of the convolutional neural network. Component a is the input layer with one hot encoding with the column number equals 64 (number of total possible codons) and the row number equals the number of codons in the transcript. Component b is the encoder component containing a sequence of layers, each consisting of a convolutional layer, followed by a Leaky Rectified Linear Unit and average pooling layer. The number of convolution layers is determined by the gene length. Component c is a fully connected layer that combined all the outputs from the component b and has k outputs for k diseases