Deep learning analysis provides a prediction at the individual level for graph classification and regression tasks. The following are the deep-learning models we have implemented:
Deep neural network
Deep neural network (DNN) is a supervised learning algorithm that learns a function by training on a dataset of a number of dimensions. It is composed of fully-connected layers and can learn a non-linear function to approximate either classification or regression. DNN can be trained on either the adjacency matrices or the graph measures. Before training a DNN model, feature selection on the input data can be done by mutual information analysis in our pipelines.
Convolutional neural network
Convolutional neural network (CNN) is a deep learning architecture that learns to extract the feature directly from data. This kind of neural network is particularly useful for finding patterns and classifying data effectively. In our pipelines, a CNN model can be training on either the adjacency matrices or the binodal graph measures.
Graph neural network
Graph neural network (GNN) is an extension of a CNN that takes node features (e.g. degree, distance) and the adjacency matrices as input to perform classification. During the training process, the multi-head self-attention mechanism is implemented in our pipelines.