Fig. 2

Flowchart of the selection method and the deep neural network architecture. Least absolute shrinkage and selection operator (LASSO) based on five-fold cross-validation was used to select optimal genomics features. The selected genes were imported into the deep learning survival (DLS) model as eigenvectors. The DLS contains multiple hidden layers, weight-decay regularization, rectifying linear units, batch normalization, dropout, and stochastic gradient descent using Nesterov momentum, gradient pruning, and learning rate scheduling. The network output is a single node that estimates the weight of the risk function parameterized through the network. IO, immunotherapy