Machine learning model for drug toxicity prediction using Graph-based neural networks

this is a small graph-based model predicting substance toxicity

The Model

the Model is make up of 2 Graph convolutional layers and two linear layers. it works by turning a molecule's chemical representation into a 1-hot encoded matrix (whuch represents atoms) and an edge matrix (which represents bonds)

Dataset

I used tox21 dataset from deepchem to train the model

Training the model

for training the model you can you can run train.py or download the pre-trained model from the release

Running the model

run test_model_cli.py

python test_model_cli.py

now enter SMILES representation of the substance:

Enter a SMILES string: CCN(CC)CCOC(=O)C1(C2CCCCC2)CCCCC1
Output: 52.33239531517029% chance of stress response to ATAD5.