DSQ

Environment

conda env create -f environment.yml
conda activate dsq-env

Download data and weights

Download model weights, ESM2 features, and data from https://drive.google.com/drive/folders/1LGd2RU1wPd1HUEOpnAN7h1mcTjEw2xgu?usp=drive_link.

The final folder tree:

|--data_module
|    |--gpcrdataset.py
|--extract_esm_feat.py
|--environment.yml
|--models
|   |--dsqgpcr.py
|--main.py
|--utils.py
|--results
|   |--dsq
|   |   |--best_checkpoint
|   |   |   |--config.json
|   |   |   |--pytorch_model.bin
|--test.py
|--data
|   |--train.csv
|   |--val.csv
|   |--test.csv
|   |--esm2_feat
|   |   |--data.mdb
|--README.md

Test

python test.py \
    --test_data_path=data/test.csv \
    --model=dsqgpcr \
    --ckpt=results/$model_name/best_checkpoint \
    --mwc_gate

Train

model_name=dsq
mkdir -p results/$model_name
python main.py \
    --train_data_path=data/train.csv \
    --val_data_path=data/val.csv \
    --output_dir=results/$model_name \
    --num_epochs=100 \
    --batch_size=2048 \
    --run_name=$model_name \
    --model=dsqgpcr \
    --patience=5 \
    --mwc_gate \
    --learning_rate 0.0005  2>&1 | tee results/$model_name/train.log