Classification of enzymes using Graph Convolutional Networks

Dataset Description -

The dataset can be loaded using pytorch_geometric dataset class.
The dataset contains the following files:

1. ENZYMES
   graph
   indicators - column vector of graph identifiers for all nodes of all graphs, the value in the i-th line is the graph_id of the node with node_id i
2. ENZYMEZ_Graph_
   labels – class labels for all graphs in the dataset, the value in the i-th line is the class label of the graph with graph_id i
3. ENZYMES_A - sparse (block diagonal) adjacency matrix for all graphs, each line corresponds to (row, col) resp. (node_id, node_id)
4. ENZYMES_node_attributes - matrix of node attributes, the comma separated values in the i-th line is the attribute vector of the node with node_id i
5. Readme file

The number of data samples included in the dataset is 600.
The label of this dataset is the 6 enzyme classes.

Problem Statement -

The dataset is being used to identify which class of enzyme a particular graph belongs to.

Models -

GCN 1 Layer GCN 2 Layers GCN 3 Layers GCN 4 Layers

Dataset Preprocessing -

I have used the NormalizeFeatures transformation to normalize the feature vectors.
I have also used stratification while splitting the dataset to ensure there is no imbalance in the proportion of classes in the training and test sets.

Training and Test Split -

I have split the dataset into 80% Train, 10% Validation and 10% Test sets.

Best Performed Models -

The GCN 4 Layer model with a Learning Rate of 0.001 performs the best out of all the models.