graphB³ is a Graph Convolutional Network (GCN)-based model designed to predict the Blood-Brain Barrier (BBB) permeability of chemical compounds. In addition to predictions, graphB³ highlights the substructures within a compound that contribute to its permeability status (BBB+ or BBB-) using GNNExplainer.

• BBB+ (permeable) substructures are highlighted in red.
• BBB- (non-permeable) substructures are highlighted in blue.

• Accurate BBB permeability prediction using graph neural networks.
• Explainability via GNNExplainer to highlight contributing substructures.
• Simple CLI usage for batch predictions.

graphB³ has been tested on Ubuntu 22.04.5 LTS.

Ensure the following packages are installed:

pip install numpy pandas matplotlib torch torchmetrics torch_geometric mxnet deepchem rdkit

Verified versions:

• numpy: 1.26.4
• pandas: 1.5.3
• logging: 0.5.1.2
• matplotlib: 3.8.2
• rdkit: 2022.09.5
• torch: 2.2.0+cu121
• mxnet: 1.9.1
• torchmetrics: 1.3.0.post0
• torch_geometric: 2.5.0

To use graphB³, download and place the following files in the same directory:

1. deepchem_utils.py - Feature extraction from SMILES input.
2. best_mcc_model_weights.pth - Pretrained graphB³ model weights.
3. graphb³.py - Main script for running graphB³.
4. bbb_test_smiles_sample.csv - Input file containing SMILES data.

Use the following command to run graphB³:

python graphb3.py \
    --input_csv /path/to/bbb_test_smiles_sample.csv \
    --model_weights /path/to/best_mcc_model_weights.pth \
    --results_csv /path/to/results/bbb_example_results.csv \
    --output_dir /path/to/results \
    > output.log &

• Input molecules must be in SMILES format and stored in a .csv file.
• The progress can be monitored via output.log.

• Prediction results will be saved in bbb_example_results.csv.
• Highlighted substructures will be stored in the GraphB3_figures/ directory.

• The data folder contains the BBB smiles data, which was used to train the model for graphB³

• Here is the link for the webserver: https://webs.iiitd.edu.in/dhanjal/graphB3/

If you would like to contribute to graphB³, feel free to submit an issue or a pull request.

If you use graphB³ in your research, please cite the relevant paper (paper link and citation).

For questions or issues, please open a GitHub issue or contact the maintainers.