pmnet_appl

Pre-trained Docking Proxy Models

Easy-to-use docking score prediction models.

Implementation List:

• TacoGFN: Target-conditioned GFlowNet for Structure-based Drug Design [paper]

If you use this implementation, please cite PharmacoNet with related papers:

Install

To use the pre-trained proxy model, you need to install torch geometric and associated libraries. You can simply install them with following scheme at the root directory:

# Case 1. Install both PharmacoNet and torch-geometric
pip install -e '.[appl]' --find-links https://data.pyg.org/whl/torch-2.3.1+cu121.html
# Case 2. Install only PharmacoNet (already torch-geometric is installed)
pip install -e .
# Case 3. In your project (already torch-geometric is installed)
pip install pharmaconet @ git+https://github.com/SeonghwanSeo/PharmacoNet.git

Load Pretrained Model

from pmnet_appl import get_docking_proxy
from pmnet_appl.tacogfn_reward import TacoGFN_Proxy

device: str | torch.device = "cuda" | "cpu"

# Cache for CrossDocked2020 Targets: 15,201 training pockets + 100 test pockets
cache_db = "train" | "test" | "all" | None

# TacoGFN Reward Function
train_dataset = "ZINCDock15M" | "CrossDocked2020"
proxy: TacoGFN_Proxy = get_docking_proxy("TacoGFN_Reward", "QVina", train_dataset, cache_db, device)
proxy = TacoGFN_Proxy.load("QVina", train_dataset, cache_db, device)

# if cache_db is 'test' | 'all'
print(proxy.scoring("14gs_A", "c1ccccc1"))
print(proxy.scoring_list("14gs_A", ["c1ccccc1", "C1CCCCC1"]))

Use custom target cache

proxy = get_docking_proxy("TacoGFN_Reward", "QVina", "ZINCDock15M", None, device)
save_cache_path = "<custom-cache-pt>"
protein_info_dict = {
    "<keyA>": ("<proteinA-pdb>", "<ref-ligand-path>"), # use center of reference ligand
    "<keyB>": ("<proteinB-pdb>", (1.0, 2.0, 3.0)),     # use center coordinates
}
proxy.get_cache_database(protein_info_dict, save_cache_path, verbose=False)

# Load Custom Target Cache
proxy = get_docking_proxy("TacoGFN_Reward", "QVina", "ZINCDock15M", save_cache_path, device)
proxy.scoring("key1", "c1ccccc1")
proxy.scoring_list("key2", ["c1ccccc1", "C1CCCCC1"])