GCLink: A Graph Contrastive Link Prediction Framework for Gene Regulatory Network Inference

Dependencies Used in This Project

• Python 3.8.0
• PyTorch 1.12.1
• torch-scatter 2.1.0
• torch-sparse 0.6.15
• torch-geometric 2.3.1

Usage

1. Set Up Conda Environment

Before running any scripts, set up the required environment using the provided environment.yml file. Create the Conda Environment:

• Navigate to the project directory and run the following command to create the environment:

  conda env create -f environment.yml

2. Gene Regulatory Network Inference

1. Preparing for gene expression profiles and TF-Target interaction files:

   • The datasets used in this project are provided in the Benchmark Datasets folder.
   • BL--ExpressionData.csv contains the gene expression matrix.
   • BL--network.csv contains the TF-Target interactions.
   • In order to use your own data, you have to provide a network file and gene expression matrix file similar to the two files mentioned above.

2. Generate Training, Validation and Testing sets:

   • First, determine the network type (Specific, Non-Specific, or STRING) and cell type to be used.
   • Then, run the following command to generate the training, validation, and testing sets:

     python Train_Test_Split.py --net <network_type> --data <cell_type>

     Replace <network_type> with the chosen network type and <cell_type> with the chosen cell type.
   • The files in folder Data/Specific/mESC 1000/sample1 provide an example of the generated data for the mESC Specific network.

3. Run Inference:

   • Set specific cell_type and sample to infer. For example:

     python GCLink_main.py -cell_type mESC -sample sample1

   • The output of the code includes:
     • AUROC and AUPRC scores for evaluation.
     • Regulatory scores for all test samples.

3. Transfer Learning

1. Training a source model
   • Train the source model using the previously generated training, validation, and testing data. Run the following command:

     python train_source.py

   • We default to using the dataset with the most available labels, the mESC Specific Network (TFs+1000).
   • The trained source model will be stored in the model folder.
2. Transfer to a target dataset
   • Select a target dataset (e.g., hESC) and run the following command:

     python transfer.py --cell_type hESC --sample sample1

   • The target dataset is generated in the same way as mentioned earlier. The example target dataset used is the hESC Specific dataset.
   • The output of the code includes:
     • AUROC and AUPRC scores for evaluation.
     • Regulatory scores for all test samples.

Contact

For any inquiries, feel free to raise issues or contact me via email at [email protected].