1. Prepare pre-training data

download data

zenodo: https://zenodo.org/doi/10.5281/zenodo.11070823 rename the folder as benchmark_data

split data

cd PPI_split
python calculate_interface_embedding.py

merge data

run ./extra_data/integrate.ipynb

2. PPBind-3D(Updating in progress）

cd PPBind
# process data
python 1-0.preprocess_PPBind3D-dataset.py \
    --summary_filepath ../PPBind-3D_example_data.csv \
    --cache_dir ./cache_data/PPBind-3D_example_data/ \
    --save_dir ./cache_data/PPBind-3D_example_data/pt/
# train
python 1-1.train.py --config ./configs/train_PPBind-3D.yml --num_workers 4

3. Prepare data for fine-tuning

cd finetune_data_DIPS-Plus

Download DIPS-Plus

wget https://zenodo.org/records/5134732/files/final_raw_dips.tar.gz
tar -xzf final_raw_dips.tar.gz

Get Batch PDB Files Downloads with Shell Script

wget https://www.rcsb.org/scripts/batch_download.sh

process data

python ./get_pair_chain_from_dill.py

This step will generate two files:
- DIPS-Plus-Pair-Data.csv
- DIPS-Plus-PDB-Set.txt

Download PDB from RCSB

mkdir ./DIPS-Plus-PDB
bash ./batch_download.sh -f ./DIPS-Plus-PDB-Set.txt -p -o ./DIPS-Plus-PDB
gunzip *.gz

Merge pretrain dataset and finetune dataset

python ./finetune_dataset_integrate.py

4. PPBind-1D(Updating in progress）

cd PPBind
# process data
python 2-0.preprocess_PPBind1D-dataset.py \
    --summary_filepath ../PPBind-1D_example_data.csv \
    --cache_dir ./cache_data/PPBind-1D_example_data/ \
    --save_dir ./cache_data/PPBind-1D_example_data/pt/
# train
python 2-1.train.py --config ./configs/train_PPBind-1D.yml --num_workers 4

5. Trained Model

You can download trained weight from Hugging Face