SF-RX

This repository contains the code used to reproduce the results from our research paper "SF-Rx: A Multi-output Deep Neural Network-Based Framework Predicting Drug-Drug Interaction under Realistic Conditions for Safe Prescription". The software is organized into four folders, each corresponding to a specific task discussed in the paper. Below are detailed instructions and notes about the code structure and data.

Folder Structure

• [SF_RX_MODEL]: Code and models for the SF-RX implementation, optimized for GPU environments.
• [GNNs]: Code for training GNNs and transformer models used in the paper.
• [FEDERATED_LEARNING]: Federated learning experiments with GPU parallelism.
• [PERMUTATION_TEST]: Permutation test for distributional shifts of scaffold structures.

Data

• All required data is located in the data folder within each directory.
• For large files, Google Drive links are provided in the respective folders.
• Note: The original results in the paper were generated using proprietary DrugBank data, which cannot be shared. Instead, we created toy datasets by combining publicly available data from DrugBank and PDR.

⚠️Note⚠️: Drugs.com data was used with retrospective permission from Drugs.com. Original source data are not publicly available!

Key Features

GPU Optimization

• SF-RX Model and Federated Learning tasks are designed to run on GPU environments.
• Federated Learning assumes 4 GPUs for parallel execution due to the computationally intensive nature of the FL experiments.
• To modify GPU settings, update the parallelism section in FEDERATED_LEARNING/experiment.py.

Dependencies

• All software and library version requirements are listed in dependencies.txt.

For any questions or issues, feel free to reach out to us via [shbae@drnoahbiotech.com], [dekim@drnoahbiotech.com], [jhyu@drnoahbiotech.com].