GARD (Graph Autoencoder Recommender for Drugs) is a medication recommendation model based on graph autoencoders. This project utilizes dimensionality reduction and graph representation learning techniques to predict drug response. By incorporating clinical data and information on drug interactions, GARD aims to provide personalized recommendations for patients in healthcare settings.
For further inquiries, please contact Mongi Kourchid at: mongi.kourchid@gmail.com.
This work, titled "Towards a Better Prescription: Graph Autoencoder for Drug Recommendation", has been published in the ICODAI 2024 conference, held in Tunis, Tunisia (https://icodai.rnu.tn/)
Ce référentiel fournit l'implémentation officielle de PyTorch du document suivant :
- Graph Autoencoder (GCN AE): Leverages advanced graph neural networks to create high-quality embeddings, enhancing the model's ability to capture complex relationships within clinical and drug interaction data.
- Accurate Drug Response Prediction: Uses learned embeddings to provide precise predictions for drug response, enabling more effective patient care and minimizing the risk of adverse reactions.
- Advanced Medication Recommendation System: Employs convolutional neural networks and deep learning methods to generate personalized drug recommendations, tailored to individual patient profiles and health conditions.
To set up and run GARD, follow these steps:
git clone https://github.com/mongikourchid/Graph-Autoencoder-Drug-Recommender-GADR.git cd Graph-Autoencoder-Drug-Recommender-GADR
python3 -m venv venv source venv/bin/activate
python -m venv venv venv\Scripts\activate
3 Install Dependencies: Ensure Python 3.7 or higher is installed. Then install the required packages from the requirements.txt file:
pip install -r requirements.txt
python main.py
python -m unittest discover tests/
The GARD model requires specific input data to train and make predictions for drug recommendations. Below is a description of the expected data formats and the steps to prepare them.
GARD uses graph-based data, where each drug is represented as a node, and interactions between drugs are represented as edges. The input data must be in a graph format.
Each drug is represented by a feature vector. These features could include molecular descriptors, drug-target interactions, or any other relevant chemical or biological features.
The relationships between drugs are represented as edges, where the edge features can include interactions like binding affinity, chemical similarity, or other types of drug-drug relationships.
A graph construction method needs to be defined, such as creating a drug interaction graph or a molecular similarity graph.
The model expects input in the form of:
Drug feature matrix (e.g., a CSV or numpy array of shape (num_drugs, num_features)). Graph adjacency matrix (e.g., a sparse matrix or a dictionary representing edges between drugs).
The GARD model generates output data that includes the predicted drug recommendations and their associated confidence scores. This output can be used for evaluation or to make recommendations.
The primary output of the model is a set of recommended drugs for a given input drug. These predictions are generated based on the learned drug representations (embeddings) and the drug interaction graph.