This repo contains gnn pipeline for submission to TDC benchmark challenge.
Use conda to install all the required dependencies. To install conda, follows these steps.
curl -LO https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
Use the env.yml file to populate the environment as follows.
conda env create -f env.yml
conda activate tdc-task
pip install git+https://github.com/bp-kelley/descriptastorus
Note that you need to ensure pyg version is 2.0.2. The results might not be reproducible in other versions.
The results are summarized in the Table below. Also refer to the report for more details.
We conducted all our experiments in AWS c5.12xlarge instances (CPU only). To reproduce the results for a specific dataset, use the following script.
python run.py --dataset hERG --verbose False