GCN DockingScalability

Our results are in the exp directory.

Build an environment

Docker

1. Create the Docker image docker build -t your_image_name .
2. Run your image docker run -it -d --shm-size 128g your_image_name
3. Access to your container docker exec -it <container id> bash

Conda

1. Create the conda env conda env create -f environment_gpu.yml
2. Activate the env conda activate conda_env_GCN_SCAL
3. Install Pyg-Lib :

pyg-lib provides efficient GPU-based routines to parallelize workloads in heterogeneous graphs across different node types and edge types.

pip install pyg-lib -f https://data.pyg.org/whl/torch-1.12.0+cu113.html

Build the Graphs

1. Download the PDBBind database from http://www.pdbbind.org.cn/ with scripts/download_pdbbind.sh. Extracted PDBBind complexes are stored in data/raw/
2. Create the graphs python data.py

Launch the trainings

1. You can set the exp/commands_list.txt to choose which trainings will be done.
2. Set launch_training.sh (lines 2 and 3) with the correct numbers
3. Run ./launch_training.sh