Context-specific Network Embedding via Contextualized Graph Attention

Installation

conda create -n cone python=3.9 -y && conda activate cone

# Upgrade pip
pip install pip setuptools -U

# Install CUDA enabled packages for CUDA 11.8 (adjust to your system accordingly)
pip install torch==2.1.0 --index-url https://download.pytorch.org/whl/cu118
pip install torch_geometric==2.4.0 torch_cluster==1.6.3 -f https://data.pyg.org/whl/torch-2.1.0+cu118.html

# Install CONE along with its requirements (in editable mode)
pip install -e .

# Optional steps
pip install -r requirements.txt  # install packages with pinned versioned
conda clean --all -y  # clean up conda environment

Usage notes

Embedding training

Run CONE embedding training for PINPPI network, using GTEx tissue expressed genes as contexts:

python main.py network=pinppi context=tissue_gtex_expr

After the training is completed, the results can be found under the dump/ directory in the run directory. By default, this will be outputs/cone-pinppi-tissue_gtex_expr-default/dump.

Evaluation

DisGeNET disease gene prediction benchmark

Run the DisGeNET evaluation on the generated embeddings:

python evaluate_disgenet.py --mode cone --emb_dir outputs/cone-pinppi-tissue_gtex_expr-default/dump/

The results will be saved to results/cone-pinppi-tissue_gtex_expr-default-disgenet.csv

PINNACLE therapeutic area prediction benchmark

Run the DisGeNET evaluation on the generated embeddings:

python evaluate_ibd_ra.py --mode cone --subset_pinnacle_genes \
    --emb_dir outputs/cone-pinppi-celltype_pinnacle-default/dump/

NOTE: must use the PINNACLE cell type context specific embeddings when --subset_pinnacle_genes is set (reproducing the setting from the original PINNACLE paper).

The results will be saved to results/cone-pinppi-celltype_pinnacle-default-pinnacle_drug_targets.csv

Cite our work

@article{liu2023cone,
  title={CONE: COntext-specific Network Embedding via Contextualized Graph Attention},
  author={Liu, Renming and Yuan, Hao and Johnson, Kayla A and Krishnan, Arjun},
  journal={bioRxiv},
  pages={2023--10},
  year={2023},
  publisher={Cold Spring Harbor Laboratory}
}

krishnanlab / cone