cheapvs_llm / vina_gpu.py
sangttruong's picture
add data
27dc10f
import os
import subprocess
import tempfile
from typing import List, Tuple
import numpy as np
import rdkit.Chem as Chem
from meeko import MoleculePreparation, PDBQTMolecule, PDBQTWriterLegacy, RDKitMolCreate
from rdkit import RDLogger
from rdkit.Chem import rdDistGeom
from dataclasses import dataclass, field
from typing import List, Optional
class StrictDataClass:
"""
A dataclass that raises an error if any field is created outside of the __init__ method.
"""
def __setattr__(self, name, value):
if hasattr(self, name) or name in self.__annotations__:
super().__setattr__(name, value)
else:
raise AttributeError(
f"'{type(self).__name__}' object has no attribute '{name}'."
f" '{type(self).__name__}' is a StrictDataClass object."
f" Attributes can only be defined in the class definition."
)
@dataclass
class VinaConfig(StrictDataClass):
opencl_binary_path: str = (
"/lfs/skampere1/0/sttruong/cheapvs_llm/Vina-GPU-2.1/AutoDock-Vina-GPU-2.1" # needed if you use VINA for rewards
)
vina_path: str = (
"/lfs/skampere1/0/sttruong/cheapvs_llm/Vina-GPU-2.1/AutoDock-Vina-GPU-2.1/AutoDock-Vina-GPU-2-1" # path to VINA executable, needed if you use VINA for rewards
)
target: str = "kras" # kras, 2bm2
VINA = '/lfs/skampere1/0/sttruong/cheapvs_llm/Vina-GPU-2.1/AutoDock-Vina-GPU-2.1/AutoDock-Vina-GPU-2-1'
logger = RDLogger.logger()
RDLogger.DisableLog("rdApp.*")
def gpu_vina_installed(vina_path=VINA):
if os.path.exists(vina_path):
return True
return False
def read_pdbqt(fn):
"""
Read a pdbqt file and return the RDKit molecule object.
Args:
- fn (str): Path to the pdbqt file.
Returns:
- mol (rdkit.Chem.rdchem.Mol): RDKit molecule object.
"""
pdbqt_mol = PDBQTMolecule.from_file(fn, is_dlg=False, skip_typing=True)
rdkitmol_list = RDKitMolCreate.from_pdbqt_mol(pdbqt_mol)
return rdkitmol_list[0]
def smile_to_conf(smile: str, n_tries=5) -> Chem.Mol:
mol = Chem.MolFromSmiles(smile)
if mol is None:
return None
Chem.SanitizeMol(mol)
tries = 0
while tries < n_tries:
params = rdDistGeom.ETKDGv3()
# set the parameters
params.useSmallRingTorsions = True
params.randomSeed = 0
params.numThreads = 1
# generate the conformer
rdDistGeom.EmbedMolecule(mol, params)
# add hydrogens
mol = Chem.AddHs(mol, addCoords=True)
if mol.GetNumConformers() > 0:
return mol
tries += 1
print(f"Failed to generate conformer for {smile}")
return mol
def mol_to_pdbqt(mol: Chem.Mol, pdbqt_file: str):
preparator = MoleculePreparation()
mol_setups = preparator.prepare(mol)
for setup in mol_setups:
pdbqt_string, is_ok, error_msg = PDBQTWriterLegacy.write_string(setup)
if is_ok:
with open(pdbqt_file, "w") as f:
f.write(pdbqt_string)
break
else:
print(f"Failed to write pdbqt file: {error_msg}")
def parse_affinty_from_pdbqt(pdbqt_file: str) -> float:
with open(pdbqt_file, "r") as f:
lines = f.readlines()
for line in lines:
if "REMARK VINA RESULT" in line:
return float(line.split()[3])
return None
script_dir = os.path.dirname(os.path.abspath(__file__))
repo_root = os.path.dirname(script_dir)
DATA_DIR = os.path.join(repo_root, "data/docking/")
TARGETS = {
"2bm2": {
"receptor": os.path.join(DATA_DIR, "2bm2/2bm2_protein.pdbqt"),
"center_x": 40.415,
"center_y": 110.986,
"center_z": 82.673,
"size_x": 30,
"size_y": 30,
"size_z": 30,
"num_atoms": 30,
},
"kras": {
"receptor": os.path.join(DATA_DIR, "kras/8azr.pdbqt"),
"ref_ligand": os.path.join(DATA_DIR, "kras/8azr_ref_ligand.sdf"),
"center_x": 21.466,
"center_y": -0.650,
"center_z": 5.028,
"size_x": 18,
"size_y": 18,
"size_z": 18,
"num_atoms": 32,
},
"trmd": {
"receptor": os.path.join(DATA_DIR, "trmd/6qrd.pdbqt"),
"center_x": 16.957,
"center_y": 21.772,
"center_z": 33.296,
"size_x": 30,
"size_y": 30,
"size_z": 30,
"num_atoms": 34,
},
'ACES_8dt5': {
'center_x': 64.615,
'center_y': 147.253,
'center_z': 6.996,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/ACES_8dt5.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/ACES_8dt5.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'ACE_1uze': {
'center_x': 40.638,
'center_y': 35.471,
'center_z': 46.563,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/ACE_1uze.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/ACE_1uze.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'ADRB2_4ldo': {
'center_x': -1.346,
'center_y': -12.351,
'center_z': -48.586,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/ADRB2_4ldo.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/ADRB2_4ldo.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'AOFB_2c66': {
'center_x': 52.765,
'center_y': 154.112,
'center_z': 26.269,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/AOFB_2c66.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/AOFB_2c66.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'BCL2_6o0k': {
'center_x': -15.357,
'center_y': 2.24,
'center_z': -9.562,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/BCL2_6o0k.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/BCL2_6o0k.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'CAH2_5gmm': {
'center_x': -0.223,
'center_y': 6.554,
'center_z': -47.066,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/CAH2_5gmm.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/CAH2_5gmm.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'CLTR1_6rz5': {
'center_x': 12.835,
'center_y': 8.127,
'center_z': -13.658,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/CLTR1_6rz5.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/CLTR1_6rz5.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'DHPS_5u14': {
'center_x': 70.928,
'center_y': -0.17,
'center_z': 101.575,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/DHPS_5u14.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/DHPS_5u14.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'DPP4_5y7k': {
'center_x': 95.486,
'center_y': -16.719,
'center_z': 60.594,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/DPP4_5y7k.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/DPP4_5y7k.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'DYR_2w3a': {
'center_x': 1.914,
'center_y': 30.495,
'center_z': -2.884,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/DYR_2w3a.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/DYR_2w3a.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'HDAC1_8voj': {
'center_x': 171.994,
'center_y': 205.147,
'center_z': 153.834,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/HDAC1_8voj.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/HDAC1_8voj.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'HMDH_1hwk': {
'center_x': 18.313,
'center_y': 8.38,
'center_z': 15.174,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/HMDH_1hwk.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/HMDH_1hwk.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'PARP1_8hlr': {
'center_x': 11.147,
'center_y': 4.004,
'center_z': -9.104,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/PARP1_8hlr.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/PARP1_8hlr.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'PBPA_3udi': {
'center_x': 34.097,
'center_y': -0.694,
'center_z': 12.515,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/PBPA_3udi.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/PBPA_3udi.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'PDE4A_3tvx': {
'center_x': 43.495,
'center_y': 16.644,
'center_z': -24.574,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/PDE4A_3tvx.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/PDE4A_3tvx.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'PPARA_7bpz': {
'center_x': 22.031,
'center_y': 0.986,
'center_z': 62.494,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/PPARA_7bpz.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/PPARA_7bpz.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'PPARG_5ugm': {
'center_x': 25.979,
'center_y': 64.86,
'center_z': -29.332,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/PPARG_5ugm.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/PPARG_5ugm.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'SC6A4_6awp': {
'center_x': 33.292,
'center_y': 185.551,
'center_z': 143.179,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/SC6A4_6awp.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/SC6A4_6awp.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
'SRC_4mxo': {
'center_x': 12.089,
'center_y': -37.176,
'center_z': -6.818,
'receptor': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/proteins_pdbqt/SRC_4mxo.pdbqt',
'ref_ligand': '/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/crystal_ligand/SRC_4mxo.pdb',
'size_x': 25.0,
'size_y': 25.0,
'size_z': 25.0,
"num_atoms": 34},
}
class QuickVina2GPU(object):
def __init__(
self,
vina_path: str = VINA,
target: str = None,
target_pdbqt: str = None,
reference_ligand: str = None,
input_dir: str = None,
out_dir: str = None,
save_confs: bool = False,
reward_scale_max: float = -1.0,
reward_scale_min: float = -10.0,
thread: int = 8000,
print_time: bool = False,
print_logs: bool = False,
):
"""
Initializes the QuickVina2GPU class with configuration for running QuickVina 2 on GPU.
Give either a code for a target or a PDBQT file.
Args:
- vina_path (str): Path to the Vina executable.
- target (str, optional): Target identifier. Defaults to None.
- target_pdbqt (str, optional): Path to the target PDBQT file. Defaults to None.
- reference_ligand (str, optional): Path to the reference ligand file. Defaults to None.
- input_dir (str, optional): Directory for input files. Defaults to a temporary directory.
- out_dir (str, optional): Directory for output files. Defaults to None, will use input_dir + '_out'.
- save_confs (bool, optional): Whether to save conformations. Defaults to False.
- reward_scale_max (float, optional): Maximum reward scale. Defaults to -1.0.
- reward_scale_min (float, optional): Minimum reward scale. Defaults to -10.0.
- thread (int, optional): Number of threads to use. Defaults to 8000.
- print_time (bool, optional): Whether to print execution time. Defaults to True.
Raises:
- ValueError: If the target is unknown.
"""
self.vina_path = vina_path
self.save_confs = save_confs
self.thread = thread
self.print_time = print_time
self.print_logs = print_logs
self.reward_scale_max = reward_scale_max
self.reward_scale_min = reward_scale_min
if target is None and target_pdbqt is None:
raise ValueError("Either target or target_pdbqt must be provided")
if input_dir is None:
input_dir = '/lfs/skampere1/0/sttruong/cheapvs_llm/vina_dir'
os.makedirs(input_dir, exist_ok=True)
self.input_dir = input_dir
self.out_dir = input_dir + "_out"
if target in TARGETS:
self.target_info = TARGETS[target]
else:
raise ValueError(f"Unknown target: {target}")
for key, value in self.target_info.items():
setattr(self, key, value)
def _write_config_file(self):
config = []
config.append(f"receptor = {self.receptor}")
config.append(f"ligand_directory = {self.input_dir}")
config.append(f"opencl_binary_path = {VinaConfig.opencl_binary_path}")
config.append(f"center_x = {self.center_x}")
config.append(f"center_y = {self.center_y}")
config.append(f"center_z = {self.center_z}")
config.append(f"size_x = {self.size_x}")
config.append(f"size_y = {self.size_y}")
config.append(f"size_z = {self.size_z}")
config.append(f"thread = {self.thread}")
with open(os.path.join(self.input_dir, "config.txt"), "w") as f:
f.write("\n".join(config))
def _write_pdbqt_files(self, smiles: List[str]):
# Convert smiles to mols
mols = [smile_to_conf(smile) for smile in smiles]
# Remove None
# mols = [mol for mol in mols if mol is not None]
# Write pdbqt files
for i, mol in enumerate(mols):
pdbqt_file = os.path.join(self.input_dir, f"input_{i}.pdbqt")
try:
mol_to_pdbqt(mol, pdbqt_file)
except Exception as e:
print(f"Failed to write pdbqt file: {e}")
def _teardown(self):
# Remove input files
for file in os.listdir(self.input_dir):
os.remove(os.path.join(self.input_dir, file))
os.rmdir(self.input_dir)
# Remove output files
if os.path.exists(self.out_dir):
for file in os.listdir(self.out_dir):
os.remove(os.path.join(self.out_dir, file))
os.rmdir(self.out_dir)
def _run_vina(self):
result = subprocess.run(
[self.vina_path, "--config", os.path.join(self.input_dir, "config.txt")], capture_output=True, text=True, cwd=VinaConfig.opencl_binary_path
)
if self.print_time:
print(result.stdout.split("\n")[-2])
if self.print_logs:
print(result.stdout.split("\n"))
if result.returncode != 0:
print(f"Vina failed with return code {result.returncode}")
print(result.stderr)
return False
def _parse_results(self):
results = []
failed = 0
for i in range(self.batch_size):
pdbqt_file = os.path.join(self.out_dir, f"input_{i}_out.pdbqt")
if os.path.exists(pdbqt_file):
affinity = parse_affinty_from_pdbqt(pdbqt_file)
else:
affinity = 0.0
failed += 1
results.append((affinity))
if failed > 0:
print(f"WARNING: Failed to calculate affinity for {failed}/{self.batch_size} molecules")
return results
def _parse_docked_poses(self):
poses = []
failed = 0
for i in range(self.batch_size):
pdbqt_file = os.path.join(self.out_dir, f"input_{i}_out.pdbqt")
if os.path.exists(pdbqt_file):
mol = read_pdbqt(pdbqt_file)
poses.append(mol)
else:
poses.append(None)
failed += 1
if failed > 0:
print(f"WARNING: Failed to read docked pdbqt files for {failed}/{self.batch_size} molecules")
return poses
def _check_outputs(self):
if not os.path.exists(self.out_dir):
return False
return True
def calculate_rewards(self, smiles: List[str]) -> List[Tuple[str, float]]:
self.batch_size = len(smiles)
# mols = [Chem.MolFromSmiles(smile) for smile in smiles]
# Write input files, config file and run vina
self._write_pdbqt_files(smiles)
self._write_config_file()
self._run_vina()
# Parse results
if self._check_outputs():
affinties = self._parse_results()
else:
affinties = [0.0] * self.batch_size
# Scale affinities to calculate rewards
affinties = np.array(affinties)
# print(
# f"AFFINITIES: mean={round(np.mean(affinties), 3 )}, std={round(np.std(affinties), 3)}, min={round(np.min(affinties), 3)}, max={round(np.max(affinties), 3)}"
# )
# Remove output files
self._teardown()
# return smiles, list(affinties), list(rewards)
return smiles, list(affinties)
def dock_mols(self, smiles: List[str]) -> List[Tuple[str, float]]:
self.batch_size = len(smiles)
# Write input files, config file and run vina
self._write_pdbqt_files(smiles)
self._write_config_file()
self._run_vina()
# Parse results
affinties = self._parse_results()
# Scale affinities to calculate rewards
affinties = np.array(affinties)
mols = self._parse_docked_poses()
print(
f"AFFINITIES: mean={round(np.mean(affinties), 3 )}, std={round(np.std(affinties), 3)}, min={round(np.min(affinties), 3)}, max={round(np.max(affinties), 3)}"
)
# Remove output files
self._teardown()
return mols, affinties
if __name__ == "__main__":
# test
smile = "Fc1cc2ccncc2cc1Br"
mol = smile_to_conf(smile)
pdbqt_file = "test.pdbqt"
mol_to_pdbqt(mol, pdbqt_file)
os.remove
parse_affinty_from_pdbqt(pdbqt_file)
# Test docking
vina = QuickVina2GPU(vina_path=VINA, target="DPP4_5y7k")
import pandas as pd
df = pd.read_csv('/lfs/skampere1/0/sttruong/cheapvs_llm/20_targets/drugs_filtered/DPP4_HUMAN.csv')
smiles_list = df['SMILES'].tolist()
outs = vina.calculate_rewards(smiles_list)
print(outs)