https://github.com/lbp2563/graph-classification-using-graph-convolutional-network-gcn-

This project implements a Graph Convolutional Network (GCN) using PyTorch Geometric for graph classification. The model is trained on the MUTAG dataset, which consists of chemical compounds labeled according to their mutagenicity.
https://github.com/lbp2563/graph-classification-using-graph-convolutional-network-gcn-

compound-components keras mutagen-synthesis neural-network tensorflow

Last synced: about 2 months ago
JSON representation

This project implements a Graph Convolutional Network (GCN) using PyTorch Geometric for graph classification. The model is trained on the MUTAG dataset, which consists of chemical compounds labeled according to their mutagenicity.

• Host: GitHub
• URL: https://github.com/lbp2563/graph-classification-using-graph-convolutional-network-gcn-
• Owner: Lbp2563
• License: mit
• Created: 2024-07-11T16:54:21.000Z (6 months ago)
• Default Branch: main
• Last Pushed: 2024-07-18T06:06:28.000Z (6 months ago)
• Last Synced: 2024-07-18T07:58:45.560Z (6 months ago)
• Topics: compound-components, keras, mutagen-synthesis, neural-network, tensorflow
• Language: Jupyter Notebook
• Homepage:
• Size: 13.7 KB
• Stars: 0
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Graph Classification using Graph Convolutional Network (GCN)

## Introduction

This project implements a Graph Convolutional Network (GCN) using PyTorch Geometric for graph classification. The model is trained on the MUTAG dataset, which consists of chemical compounds labeled according to their mutagenicity.

## Dataset

The MUTAG dataset is a collection of nitroaromatic compounds aimed at predicting their mutagenicity on *Salmonella typhimurium*. The dataset includes 188 samples of chemical compounds with 7 discrete node labels. The dataset files include:

1. **DS_A.txt**: Sparse (block diagonal) adjacency matrix for all graphs.

2. **DS_graph_indicator.txt**: Column vector of graph identifiers for all nodes of all graphs.

3. **DS_graph_labels.txt**: Class labels for all graphs in the dataset.

4. **DS_node_labels.txt**: Column vector of node labels.

Optional files, if available:

- **DS_edge_labels.txt**: Labels for the edges.

- **DS_edge_attributes.txt**: Attributes for the edges.

- **DS_node_attributes.txt**: Matrix of node attributes.

- **DS_graph_attributes.txt**: Regression values for all graphs in the dataset.

### Node Labels

- 0: Carbon (C)

- 1: Nitrogen (N)

- 2: Oxygen (O)

- 3: Fluorine (F)

- 4: Iodine (I)

- 5: Chlorine (Cl)

- 6: Bromine (Br)

### Edge Labels

- 0: Aromatic

- 1: Single

- 2: Double

- 3: Triple

## Model Architecture

The Graph Convolutional Network (GCN) is designed with the following components:

1. Input: Node features, edge index, and batch information.

2. Layers:

   - Three Graph Convolutional layers with ReLU activation function.

   - Global Mean Pooling layer.

   - Dropout layer (p=0.5).

   - Linear classifier.

### Algorithm

The GCN algorithm operates on graph-structured data:

1. Input node features, edge indices, and batch information.

2. Perform graph convolution using multiple GCN layers.

3. Aggregate node-level features into a fixed-size representation for each graph using global pooling.

4. Apply a linear classifier to predict the output class probabilities.

## Training

- **Optimizer**: Adam optimizer with learning rate 0.01.

- **Loss Function**: Cross-Entropy Loss.

- **Procedure**:

  - Iterate over the training dataset in batches.

  - Perform forward pass.

  - Compute loss.

  - Backpropagate gradients.

  - Update parameters.

## Testing

- Evaluate the model on both training and test datasets.

- Compute accuracy as the ratio of correct predictions to the total number of samples.

## Code Structure

1. **Data Loading**:

   - Load MUTAG dataset using PyTorch Geometric.

   - Split the dataset into training and test sets.

2. **Model Definition**:

   - Define the GCN model architecture using PyTorch Geometric.

3. **Training and Evaluation**:

   - Train the model using the training dataset.

   - Evaluate the model on both training and test datasets.

   - Display training and test accuracies for each epoch.

## Results

- **Training Accuracy**: 0.7933

- **Test Accuracy**: 0.7632

## References

Debnath, A. K., Lopez de Compadre, R. L., Debnath, G., Shusterman, A. J., & Hansch, C. (1991). Structure-activity relationship of mutagenic aromatic and heteroaromatic nitro compounds. Correlation with molecular orbital energies and hydrophobicity. *Journal of Medicinal Chemistry*, 34(2), 786-797.

## Dataset Link

[MUTAG Dataset](https://paperswithcode.com/dataset/mutag)

## Author

[Krishi Desai](https://github.com/krishi-03)

[Lakshit Pathak](https://github.com/Lakshit-25)

[Lakshin Pathak](https://github.com/Lbp2563)

## License

This project is licensed under the MIT License.