A graph neural network (GNN) model to predict protein-protein interactions (PPI) with no sample features
This is a simple graph convolutional network (GCN) to predict the protein-protein interactions. There are 2 datasets: a large one and a small one. The useful informations in the dataset are only known protein-protein interactions and the bioinformatic database query of proteins. Since crawling more informations from the database is troublesome, in this project, I want to predict PPIs with only there known interaction relationships, so GCN is utilized.
- Python 3.8.3
- PyTorch 1.6.0
Repo content explanation
- dataset folder contains two .txt files: a larger dataset and a small dataset.
train.pyis the script defining the GCN model and training it.
metrics.py: compute the metrics for performance evaluation.
topInteract.py: choose the protein-protein pairs with highest score as the predicted PPIs.
- *_files folders contain the output files when training based on different dataset.