Genen

Genen is a project that utilizes Graph Neural Networks (GNNs) to analyze PPI, GO, and GO+PPI networks to predict various gene attributes such as solubility. The project compares the performance of different networks and models in predicting these attributes.

Project Structure

• PPI.ipynb: Uses GCN and GAT to predict gene solubility in the PPI network.

• GO.ipynb: Uses GCN and GAT to predict gene solubility in the GO network.

• GO+PPI.ipynb: Uses GCN and GAT to predict gene solubility in the GO+PPI network and includes hidden node feature testing.

• regression.ipynb: Uses GCN for regression to predict gene conservation.

• PPI_hidding.ipynb: Tests hidden node features in the PPI network.

• other.ipynb: Uses GCN to predict other gene attributes, including:

  • Dosage Sensitivity
  • BivalentVs Lys4 Methylated
  • BivalentVs Non Methylated
  • Tf range
  • Tf target type
  • Solubility
  • Subcellular localization

• GNN/: Directory containing all the datasets used in this project.

• GCN.pth and GAT.pth: Model parameters trained on the GO+PPI network.

Results

The prediction accuracy on the GO+PPI network for solubility is generally higher compared to the PPI and GO networks. The trained GCN and GAT network parameters are saved in the GCN.pth and GAT.pth files.

Data

All data files are stored in the google drive. Ensure that the data is correctly placed before running the notebooks.