DABI-DDI

Description of the Code Execution

All experimental codes for the Bayes model can be executed directly by running the bayes_train.py file located in the Bayes folder. By executing this file, the posterior probabilities of adverse events derived from the FAERS dataset using the Bayesian model will be obtained. Additionally, a comparison plot of AUC curves between the Bayesian model and the signal detection algorithm is generated. The output also includes the probabilities of adverse events associated with various drug combinations and their corresponding confidence levels. This enables the identification of drug combinations with a high confidence of causing adverse events through the Bayesian approach.

For all experiments related to the Mymodel model, the dataset can first be created using the create_data.py file in the Mymodel folder. The model training and validation process can then be performed by running the mymodel_train.py file. All experimental results are saved in the result folder, including the comparative performance of the proposed model against other models, ablation study results, and ROC curve comparison plots. The biological attribute network can be visualized by directly executing the BAN.py file.

All data processing and model training were conducted on an Ubuntu 20.04 LTS system, and the codebase was developed using the Torch framework. The required Python libraries for running the experiments are listed in the requirement.txt file.