Relational Learning for Molecule Translation from Noisy Images

UCLA CS 267 (Spring 2022) project by Zhouxing Shi, Boyang Fu, Yifei Xu, Yuxin Wu.

We propose to improve molecule translation by leveraging the graphical structure information of molecules with Graph Neural Networks.

Data

Download generated data from Google Drive, and extract the data to the data/ folder. It is recommended to use the following commands wtih gdown which can be installed by pip:

cd data
gdown --id 1_dvhLNvONUhIAWIsfnFKQlyQEqRXIknA
gdown --id 1Ke3aaCh_HuygDMlduxtN1xOZ-h3-cFKM
gdown --id 1tF8X4bo2n_CFuj6Jd2xhwgosXUutEWW6
gdown --id 1QFMMRqkzbrNX6WIPUhFEes5CuBbOM75r
gdown --id 1zgBBqgvv9cUVWHK8NErd-0_DixSZyARN
mkdir pretrained
cd pretrained
gdown --id 1u8MoXGBnvkuqSIbYka4C1Fg_cz2mjxuQ
gdown --id 1Xj5iooTYrXAn8Hi9RFRYB03Vv746_G_w
gdown --id 1kCmp9ZNc-S-XHIVZBWS-oJa7uKpxHzd3
unzip image_data_training.zip
unzip image_data_test.zip
mv data_subset image_training
mv data_test image_test

Dependencies

Installl PyG first. Some wheels provided by PyG may be helpful if is too slow to build some dependencies locally.

Installing PyG by conda:

conda create --name pyg python=3.8
conda activate pyg
pip install torch torchvision torchaudio
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.11.0+cpu.html

CUDA version for pytorch 1.8:

pip install https://data.pyg.org/whl/torch-1.8.0%2Bcu111/torch_cluster-1.5.9-cp38-cp38-linux_x86_64.whl
pip install https://data.pyg.org/whl/torch-1.8.0%2Bcu111/torch_scatter-2.0.8-cp38-cp38-linux_x86_64.whl
pip install https://data.pyg.org/whl/torch-1.8.0%2Bcu111/torch_sparse-0.6.9-cp38-cp38-linux_x86_64.whl
pip install torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.8.0%2Bcu111.html

Install other Python libraries:

pip install -r requirements.txt

Experiments

Set a DIR environment variable to specify a prefix of directories for saving trained models.

Node Classification

python main.py --device cuda  --save-dir $DIR\_node_GCN --model-graph GCN_large --task node 
python main.py --device cuda  --save-dir $DIR\_node_GAT --model-graph GAT_large --task node
python main.py --device cuda  --save-dir $DIR\_node_GAT_no_edge_attr --model-graph GAT_large_no_edge_attr --task node

Link Classification

python main.py --device cuda  --save-dir $DIR\_link_GCN --model-graph GCN_large --task link
python main.py --device cuda  --save-dir $DIR\_link_GAT --model-graph GAT_large --task link
python main.py --device cuda  --save-dir $DIR\_link_GAT_no_edge_attr --model-graph GAT_large_no_edge_attr --task link

Running Other Commands

Run python main.py --help to see available options.

Pre-Trained Models

• Graph models