Drug-Toxicity-Prediction-Project

• Convolutional Neural Network for detecting drug toxicity
• Drug Input: Simplified Molecular-Input Line-Entry System (SMILES) Expression
• Example:

Implementation

1. The SMILES Expression of drug is transformed into one-hot encoding.
2. Model Representation: implement 4 different architecture of CNN models (using TensorFlow 1.15.4)

Model 1

Model 2

Model 3

Model 4

3. Early stopping is used in training progress.

Conclusion

• After testing, model 4 achieve the best result for predicting drug toxicity without overfitting,
• Model 4 is used to build the final version of model in final_version folder.
• Some pretrained models are saved at model folder.