Before using PharmacoNet, consider using OpenPharmaco - GUI powered by PharmacoNet.
Accepted in NeurIPS Workshop 2023 (AI4D3 | New Frontiers of AI for Drug Discovery and Development) [arxiv]
Official Github for PharmacoNet: Accelerating Large-Scale Virtual Screening by Deep Pharmacophore Modeling by Seonghwan Seo* and Woo Youn Kim.
- Fully automated protein-based pharmacophore modeling based on image instance segmentation modeling
- Coarse-grained graph matching at the pharmacophore level for high throughput
- Pharmacophore-aware scoring function with parameterized analytical function for robust generalization ability
PharmacoNet is an extremely rapid yet reasonably accurate ligand evaluation tool with high generation ability.
If you have any problems or need help with the code, please add an github issue or contact [email protected].
# Pharmacophore Modeling
python modeling.py --pdb <PDB ID> # RCSB PDB importing
python modeling.py --protein <PROTEIN_PATH> --prefix <EXP_NAME> --cuda # CUDA acceleration
python modeling.py --protein <PROTEIN_PATH> --prefix <EXP_NAME> --ref_ligand <REF_LIGAND_PATH>
# Virtual Screening
python screening.py -p <MODEL_PATH> --library <LIBRARY_DIR> --out <RESULT_PATH> --cpus <NCPU>
# Feature Extraction for Deep Learning Researcher
python feature_extraction.py --protein <PROTEIN_PATH> --ref_ligand <REF_LIGAND_PATH> --out <SAVE_PKL_PATH>
python feature_extraction.py --protein <PROTEIN_PATH> --center <X> <Y> <Z> --out <SAVE_PKL_PATH> --cudaFor various environment including Linux, MacOS and Window, the script installs cpu-only version of PyTorch by default. You can install a cuda-available version by modifying environment.yml or installing PyTorch manually.
conda create -f environment.yml
conda activate pmnet
pip install torch # 1.13<=torch<=2.3.1, CUDA acceleration is available. 1min for 1 cpu, 10s for 1 gpu
pip install .# Required python>=3.9, Best Performance at higher version. (3.9, 3.10, 3.11, 3.12 - best)
conda create --name openph python=3.10 openbabel=3.1.1 pymol-open-source=3.0.0 numpy=1.26.4
conda activate pmnet
pip install torch # 1.13<=torch<=2.3.1, CUDA acceleration is available. 1min for 1 cpu, 10s for 1 gpu
pip install rdkit biopython omegaconf tdqm numba # Numba is optional, but recommended.
pip install molvoxel # Molecular voxelization tools with minimal dependencies (https://github.com/SeonghwanSeo/molvoxel.git)You can run model.py for automated protein-based pharmacophore modeling with RCSB PDB code or custom protein path (--protein). With protein path, you should enter --prefix.
The pharmacophore model file is result/6oim/6oim_D_MOV_model.pm and the pymol session file is result/6oim/6oim_D_MOV_model.pse
# Pharmacophore Modeling for KRAS(G12C) - PDBID: 6OIM
> python modeling.py --pdb 6oim
INFO:root:Load PharmacoNet finish
INFO:root:Download 6oim to result/6oim/6oim.pdb
==============================
INFO:root:A total of 3 ligand(s) are detected!
Ligand 1
- ID : MG (Chain: B [auth A])
- Center : -2.512, 2.588, 0.220
- Name : MAGNESIUM ION
Ligand 2
- ID : GDP (Chain: C [auth A])
- Center : -6.125, 3.588, 7.310
- Name : GUANOSINE-5-DIPHOSPHATE
Ligand 3
- ID : MOV (Chain: D [auth A])
- Center : 1.872, -8.260, -1.361
- Name : AMG 510 (BOUND FORM)
- Synonyms: 6-FLUORO-7-(2-FLUORO-6-HYDROXYPHENYL)-4-[(2S)-2-METHYL-4-PROPANOYLPIPERAZIN-1-YL]-1-[4-METHYL-2-(PROPAN-2-YL)PYRIDIN-3-YL]PYRIDO[2,3-D]PYRIMIDIN-2(1H)-ONE
INFO:root:Select the ligand number(s) (ex. 3 ; 1,3 ; manual ; all ; exit)
ligand number:3 # USER INPUT: Enter the ligand number for binding site detection
INFO:root:Running 3th Ligand...
Ligand 3
- ID : MOV (Chain: D [auth A])
- Center : 1.872, -8.260, -1.361
- Name : AMG 510 (BOUND FORM)
- Synonyms: 6-FLUORO-7-(2-FLUORO-6-HYDROXYPHENYL)-4-[(2S)-2-METHYL-4-PROPANOYLPIPERAZIN-1-YL]-1-[4-METHYL-2-(PROPAN-2-YL)PYRIDIN-3-YL]PYRIDO[2,3-D]PYRIMIDIN-2(1H)-ONE
INFO:root:Save Pharmacophore Model to result/6oim/6oim_D_MOV_model.pm
INFO:root:Save Pymol Visualization Session to result/6oim/6oim_D_MOV_model.pse# With reference ligand.
> python modeling.py --protein ./examples/6OIM_protein.pdb --ref_ligand ./examples/6OIM_D_MOV.pdb --prefix 6oim
INFO:root:Load PharmacoNet finish
INFO:root:Load examples/6OIM_protein.pdb
INFO:root:Using center of examples/6oim_D_MOV.pdb as center of box
INFO:root:Save Pharmacophore Model to result/6oim/6oim_6oim_D_MOV_model.pm
INFO:root:Save Pymol Visualization Session to result/6oim/6oim_6oim_D_MOV_model.pse
# Without reference ligand -> center is required.
> python modeling.py --protein ./examples/6OIM_protein.pdb --prefix 6oim
INFO:root:Load PharmacoNet finish
INFO:root:Load examples/6OIM_protein.pdb
WARNING:root:No ligand is detected!
INFO:root:Enter the center of binding site manually:
x: 2 # USER INPUT: Enter x
y: -8 # USER INPUT: Enter y
z: -1 # USER INPUT: Enter z
INFO:root:Using center (2.0, -8.0, -1.0)
INFO:root:Save Pharmacophore Model to result/6OIM/6OIM_2.0_-8.0_-1.0_model.pm
INFO:root:Save Pymol Visualization Session to result/6OIM/6OIM_2.0_-8.0_-1.0_model.pseWe provide the simple script for screening.
# Default Parameter Setting (Cation/Anion: 8, Aromatic/Halogen/HBA/HBD: 4, Hydrophobic: 1)
python screening.py -p <MODEL_PATH> --library <LIBRARY_DIR> --out <RESULT_PATH> --cpus <NCPU>
# Custom Parameters Setting
python screening.py -p <MODEL_PATH> --library <LIBRARY_DIR> --out <RESULT_PATH> --cpus <NCPU> \
--anion <ANION> --cation <CATION> --aromatic <AROMATIC> \
--hbd <HBD> --hba <HBA> --halogen <HALOGEN> --hydrophobic <HYDROPHOBIC>
# Example
python screening.py -p ./result/6oim/6oim_D_MOV_model.pm --library examples/library --out result.csv --cpus 1
python screening.py -p ./result/6oim/6oim_D_MOV_model.pm --library examples/library --out result.csv --cpus 2 --hbd 5 --hba 5 --aromatic 8Also, it can be easily included in your custom script via the python code below. (* Multiprocessing is allowed)
from pmnet import PharmacophoreModel
model = PharmacophoreModel.load(<PHARMCOPHORE_MODEL_PATH>)
# NOTE: Scoring with ligand file with 1 or more conformers
score = model.scoring_file(<LIGAND_PATH>) # SDF, MOL2, PDB
# NOTE: Scoring with RDKit ETKDG Conformers
score = model.scoring_smiles(<SMILES>, <NUM_CONFORMERS>)For deep learning researcher who want to use PharmacoNet as pre-trained model for feature extraction, we provide the script feature_extraction.py.
python feature_extraction.py --protein <PROTEIN_PATH> --ref_ligand <REF_LIGAND_PATH> --out <SAVE_PT_PATH>
python feature_extraction.py --protein <PROTEIN_PATH> --center <X> <Y> <Z> --out <SAVE_PT_PATH>OUTPUT=(multi_scale_features, hotspot_info)
multi_scale_features: list[torch.Tensor]:
- torch.Tensor [96, 4, 4, 4]
- torch.Tensor [96, 8, 8, 8]
- torch.Tensor [96, 16, 16, 16]
- torch.Tensor [96, 32, 32, 32]
- torch.Tensor [96, 64, 64, 64]
hotspot_infos: list[hotspot_info]
info: dict[str, Any]
- hotspot_feature: torch.Tensor (192,)
- hotspot_position: tuple[float, float, float] - (x, y, z)
- hotspot_score: float in [0, 1]
- nci_type: str (10 types)
'Hydrophobic': Hydrophobic interaction
'PiStacking_P': PiStacking (Parallel)
'PiStacking_T': PiStacking (T-shaped)
'PiCation_lring': Interaction btw Protein Cation & Ligand Aromatic Ring
'PiCation_pring': Interaction btw Protein Aromatic Ring & Ligand Cation
'SaltBridge_pneg': SaltBridge btw Protein Anion & Ligand Cation
'SaltBridge_lneg': SaltBridge btw Protein Cation & Ligand Anion
'XBond': Halogen Bond
'HBond_pdon': Hydrogen Bond btw Protein Donor & Ligand Acceptor
'HBond_ldon': Hydrogen Bond btw Protein Acceptor & Ligand Donor
- hotspot_type: str (7 types)
{'Hydrophobic', 'Aromatic', 'Cation', 'Anion',
'Halogen', 'HBond_donor', 'HBond_acceptor'}
*** `type` is obtained from `nci_type`.
- point_type: str (7 types)
{'Hydrophobic', 'Aromatic', 'Cation', 'Anion',
'Halogen', 'HBond_donor', 'HBond_acceptor'}
*** `type` is obtained from `nci_type`.For feature extraction, it is recommended to use score_threshold=0.5 instead of default setting used for pharmacophore modeling. If you want to extract more features, decrease the score_threshold.
from pmnet.module import PharmacoNet, parse_protein
module = PharmacoNet(
"cuda",
score_threshold = 0.5, # <SCORE_THRESHOLD: float | dict[str, float], recommended=0.5>,
molvoxel_library = 'numpy' # <MOLVOXEL_LIBRARY: str, if you use it in `Dataset`, set 'numpy'>
)
# End-to-End calculation
multi_scale_features, hotspot_infos = module.feature_extraction(<PROTEIN_PATH>, <REF_LIGAND_PATH>)
multi_scale_features, hotspot_infos = module.feature_extraction(<PROTEIN_PATH>, center=(<X>, <Y>, <Z>))
# Step-wise calculation
voxelizer = module.voxelizer
# In Dataset (Type: Tuple[Tensor, Tensor, Tensor, Tensor])
protein_data = module.parse_protein(voxelizer, <PROTEIN_PATH>, <REF_LIGAND_PATH>, <CENTER_NOISE>)
# In Model
multi_scale_features, hotspot_infos = module.run_extraction(protein_data)- TacoGFN [paper]
Paper on arxiv
@article{seo2023pharmaconet,
title = {PharmacoNet: Accelerating Large-Scale Virtual Screening by Deep Pharmacophore Modeling},
author = {Seo, Seonghwan and Kim, Woo Youn},
journal = {arXiv preprint arXiv:2310.00681},
year = {2023},
url = {https://arxiv.org/abs/2310.00681},
}