def test_Oracle(self):
from tdc import Oracle
from tdc import Oracle
oracle = Oracle(name='SA')
x = oracle(['CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O'])
oracle = Oracle(name='Hop')
x = oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O'])
def get(self, benchmark, num_max_call=5000):
dataset = fuzzy_search(benchmark, self.dataset_names)
data_path = os.path.join(self.path, dataset)
if self.file_format == 'csv':
train = pd.read_csv(os.path.join(data_path, 'train_val.csv'))
test = pd.read_csv(os.path.join(data_path, 'test.csv'))
elif self.file_format == 'pkl':
train = pd.read_pickle(os.path.join(data_path, 'train_val.pkl'))
test = pd.read_pickle(os.path.join(data_path, 'test.pkl'))
elif self.file_format == 'oracle':
target_pdb_file = os.path.join(self.path, dataset + '.pdb')
if self.name == 'docking_group':
oracle = Oracle(name="Docking_Score",
software="vina",
pyscreener_path=self.pyscreener_path,
receptors=[target_pdb_file],
center=docking_target_info[dataset]['center'],
size=docking_target_info[dataset]['size'],
buffer=10,
path=data_path,
num_worker=self.num_workers,
ncpu=self.num_cpus,
num_max_call=num_max_call)
return {'oracle': oracle, 'name': dataset}
else:
return {'train_val': train, 'test': test, 'name': dataset}
def __init__(self, name):
## DRD2 GSK3B JNK3 cyp3a4_benchmark
from tdc import Oracle
self.name = name
super().__init__(score_modifier=None)
if 'docking' not in self.name.lower(): ### drd2 gsk3 JNK3
self.oracle = Oracle(name=self.name)
elif self.name.lower() == 'docking_5wiu':
self.oracle = Oracle(
name='Docking_Score',
software='vina',
pyscreener_path='/project/molecular_data/graphnn/pyscreener',
receptors=[
'/project/molecular_data/graphnn/pyscreener/testing_inputs/5WIU.pdb'
],
docked_ligand_file=
'/project/molecular_data/graphnn/pyscreener/testing_inputs/5WIU_with_ligand.pdb',
buffer=10,
path='/project/molecular_data/graphnn/pyscreener/my_test/',
num_worker=1,
ncpu=4)
elif self.name.lower() == 'docking_drd3':
self.oracle = Oracle(
name='Docking_Score',
software='vina',
pyscreener_path='/project/molecular_data/graphnn/pyscreener',
receptors=[
'/project/molecular_data/graphnn/pyscreener/testing_inputs/DRD3.pdb'
],
center=(9, 22.5, 26),
size=(15, 15, 15),
buffer=10,
path='/project/molecular_data/graphnn/pyscreener/my_test/',
num_worker=1,
ncpu=10)
self.docking_num_file = "/project/molecular_data/graphnn/pyscreener/docking_num.txt"
write_num(self.docking_num_file, 0)
print('----------initialize docking_num_file-------------')
def __next__(self):
if self.index < self.num_datasets:
dataset = self.dataset_names[self.index]
print_sys('--- ' + dataset + ' ---')
data_path = os.path.join(self.path, dataset)
if not os.path.exists(data_path):
os.mkdir(data_path)
if self.file_format == 'csv':
train = pd.read_csv(os.path.join(data_path, 'train_val.csv'))
test = pd.read_csv(os.path.join(data_path, 'test.csv'))
elif self.file_format == 'pkl':
train = pd.read_pickle(os.path.join(data_path,
'train_val.pkl'))
test = pd.read_pickle(os.path.join(data_path, 'test.pkl'))
elif self.file_format == 'oracle':
target_pdb_file = os.path.join(self.path, dataset + '.pdb')
self.index += 1
if self.name == 'docking_group':
oracle = Oracle(name="Docking_Score",
software="vina",
pyscreener_path=self.pyscreener_path,
receptors=[target_pdb_file],
center=docking_target_info[dataset]['center'],
size=docking_target_info[dataset]['size'],
buffer=10,
path=data_path,
num_worker=self.num_workers,
ncpu=self.num_cpus,
num_max_call=self.num_max_call)
return {'oracle': oracle, 'name': dataset}
else:
return {'train_val': train, 'test': test, 'name': dataset}
else:
raise StopIteration
from tdc import Oracle oracle = Oracle(name='isomers_c7h8n2o2') print(oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O']))
from absl import app
from absl import flags
from rdkit import Chem
from rdkit import DataStructs
from rdkit.Chem import AllChem
from tensorflow.compat.v1 import gfile
from dqn import deep_q_networks
from dqn import molecules as molecules_mdp
from dqn import run_dqn
from dqn.py import molecules
from dqn.tensorflow_core import core
from tdc import Oracle
qed = Oracle(name = 'qed')
logp = Oracle(name = 'logp')
jnk = Oracle(name = 'JNK3')
gsk = Oracle(name = 'GSK3B')
from scipy.stats import gmean
def logp_modifier(logp_score):
return max(0.0,min(1.0,1/14*(logp_score+10)))
def qed_logp_jnk_gsk_fusion(qed_score, logp_score, jsn_score, gsk_score):
logp_score = logp_modifier(logp_score)
gmean_score = gmean([qed_score, logp_score, jsn_score, gsk_score])
def test_Oracle(self):
from tdc import Oracle
oracle = Oracle(name='Hop')
print(oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O']))
def test_Oracle(self):
from tdc import Oracle
oracle = Oracle(name='celecoxib rediscovery')
print(oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O']))
from tdc import Oracle oracle = Oracle(name='aripiprazole_similarity') print(oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O']))
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--oracle_num', type=int, default=1500)
parser.add_argument('--oracle_name',
type=str,
default="qed",
choices=['jnkgsk', 'qedsajnkgsk', 'qed', 'jnk', 'gsk'])
parser.add_argument('--generations', type=int, default=50)
parser.add_argument('--population_size', type=int, default=20)
args = parser.parse_args()
oracle_num = args.oracle_num
oracle_name = args.oracle_name
generations = args.generations
population_size = args.population_size
start_smiles_lst = ['C1(N)=NC=CC=N1'] ## 'C1=CC=CC=C1NC2=NC=CC=N2'
qed = Oracle('qed')
sa = Oracle('sa')
jnk = Oracle('JNK3')
gsk = Oracle('GSK3B')
logp = Oracle('logp')
mu = 2.230044
sigma = 0.6526308
def normalize_sa(smiles):
sa_score = sa(smiles)
mod_score = np.maximum(sa_score, mu)
return np.exp(-0.5 * np.power((mod_score - mu) / sigma, 2.))
if oracle_name == 'jnkgsk':
def oracle(smiles):
return np.mean((jnk(smiles), gsk(smiles)))
elif oracle_name == 'qedsajnkgsk':
def oracle(smiles):
return np.mean(
(qed(smiles), normalize_sa(smiles), jnk(smiles), gsk(smiles)))
elif oracle_name == 'qed':
def oracle(smiles):
return qed(smiles)
elif oracle_name == 'jnk':
def oracle(smiles):
return jnk(smiles)
elif oracle_name == 'gsk':
def oracle(smiles):
return gsk(smiles)
elif oracle_name == 'logp':
def oracle(smiles):
return logp(smiles)
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
device = 'cpu' ## cpu is better
model_ckpt = "save_model/GNN_epoch_0_validloss_1.61160.ckpt"
gnn = torch.load(model_ckpt)
gnn.switch_device(device)
result_pkl = "result/" + oracle_name + ".pkl"
optimization(start_smiles_lst,
gnn,
oracle,
oracle_num,
oracle_name,
generations=generations,
population_size=population_size,
lamb=2,
topk=5,
epsilon=0.7,
result_pkl=result_pkl)
def test_Oracle(self):
# Molecule Generation Oracles
from tdc import Oracle
oracle = Oracle(name='GSK3B')
smiles_lst = [
'CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'C[C@@H]1CCc2c(sc(NC(=O)c3ccco3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O']
oracle(smiles_lst)
oracle = Oracle(name='DRD2')
smiles_lst = [
'CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'C[C@@H]1CCc2c(sc(NC(=O)c3ccco3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O']
oracle(smiles_lst)
oracle = Oracle(name='Hop')
oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O'])
oracle = Oracle(name='Valsartan_SMARTS')
oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O'])
oracle = Oracle(name='Rediscovery')
oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O'])
oracle = Oracle(name='SA')
smiles_lst = [
'CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'C[C@@H]1CCc2c(sc(NC(=O)c3ccco3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O']
oracle(smiles_lst)
oracle = Oracle(name='Uniqueness')
smiles_lst = [
'CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'C[C@@H]1CCc2c(sc(NC(=O)c3ccco3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O']
oracle(smiles_lst)
oracle = Oracle(name='Novelty')
smiles_lst = [
'CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'C[C@@H]1CCc2c(sc(NC(=O)c3ccco3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O']
oracle(smiles_lst, smiles_lst)
oracle = Oracle(name='Diversity')
smiles_lst = [
'CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'C[C@@H]1CCc2c(sc(NC(=O)c3ccco3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O']
oracle(smiles_lst)
oracle = Oracle(name='Scaffold Hop')
oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O'])
def test_Oracle(self):
from tdc import Oracle
oracle = Oracle(name='isomers_c7h8n2o2')
print(oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O']))
from tdc import Oracle
oracle = Oracle(name = 'celecoxib rediscovery')
print(oracle(['CC(=O)OC1=CC=CC=C1C(=O)O',
'C1=CC=C(C=C1)C=O']))
from tdc import Oracle
oracle = Oracle(name = 'Hop')
print(oracle(['CC(=O)OC1=CC=CC=C1C(=O)O',
'C1=CC=C(C=C1)C=O']))
def test_Oracle(self):
from tdc import Oracle
oracle = Oracle(name='aripiprazole_similarity')
print(oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O']))
from guacamol.goal_directed_generator import GoalDirectedGenerator
# from guacamol.scoring_function import ScoringFunction
from guacamol.utils.chemistry import canonicalize_list, canonicalize
from joblib import delayed
from smiles_lstm_hc.rnn_generator import SmilesRnnMoleculeGenerator
from smiles_lstm_hc.rnn_utils import load_rnn_model
from tdc import Oracle
drd3_oracle = Oracle(
name='Docking_Score',
software='vina',
pyscreener_path='/project/molecular_data/graphnn/pyscreener',
receptors=[
'/project/molecular_data/graphnn/pyscreener/testing_inputs/DRD3.pdb'
],
center=(9, 22.5, 26),
size=(15, 15, 15),
buffer=10,
path='/project/molecular_data/graphnn/pyscreener/my_test/',
num_worker=1,
ncpu=10)
global oracle_num
oracle_num = 0
def drd3_docking_oracle(smiles):
# oracle_num += 1
# print('Docking call', oracle_num)
return min(max(-drd3_oracle(smiles) / 15.0, 0), 1)
def evaluate(self,
pred,
true=None,
benchmark=None,
criteria='all',
m1_api=None):
if self.name == 'docking_group':
results_all = {}
for data_name, pred_ in pred.items():
results = {}
## pred is a list of smiles strings
if len(pred_) != 100:
raise ValueError(
"The expected output is a list of top 100 molecules!")
dataset = fuzzy_search(benchmark, self.dataset_names)
# docking scores for the top K smiles (K <= 100)
target_pdb_file = os.path.join(self.path, dataset + '.pdb')
oracle = Oracle(name="Docking_Score",
software="vina",
pyscreener_path=self.pyscreener_path,
receptors=[target_pdb_file],
center=docking_target_info[dataset]['center'],
size=docking_target_info[dataset]['size'],
buffer=10,
path=data_path,
num_worker=self.num_workers,
ncpu=self.num_cpus,
num_max_call=10000)
docking_scores = oracle(pred_)
results['docking_scores_dict'] = docking_scores
values = np.array(list(docking_scores.values()))
results['AVG_Top100'] = np.mean(values)
results['AVG_Top10'] = np.mean(sorted(values)[:10])
results['Top1'] = max(values)
all_criteria = [
'm1', 'filters', 'diversity', 'validity', 'uniqueness'
]
if criteria == 'all':
criteria = all_criteria
elif criteria == 'none':
criteria = []
else:
if sum([1 if i in all_criteria else 0
for i in criteria]) != len(criteria):
# there is at least one criteria does not match the supported evaluation
raise ValueError(
"Please select the criteria from a list of 'm1', 'filters', 'diversity', 'validity', 'uniqueness'!"
)
if 'm1' in criteria:
if m1_api is None:
raise ValueError(
"Please input the m1_api token in the evaluate function call! You can obtain it via: https://tdcommons.ai/functions/oracles/#moleculeone"
)
m1 = Oracle(name='Molecule One Synthesis',
api_token=m1_api)
m1_scores = m1(pred_)
scores_array = list(m1_scores.values())
results['m1_scores_dict'] = m1_scores
results['AVG_m1_scores'] = np.mean(scores_array)
## TODO: how good is the m1 score? ask stan; 0.5 placeholder
results['AVG_docking_scores_synthesizable'] = np.mean([
docking_scores[i] for i, j in m1_scores.items()
if j > 0.5
])
if 'filters' in criteria:
from tdc.chem_utils import MolFilter
## TODO: select an optimal set of filters. test a bit.
filters = MolFilter(filters=['PAINS'], HBD=[0, 6])
pred_filter = filters(pred_)
results['pass_filter_smiles_list'] = pred_filter
results['unfiltered_fractions'] = float(
len(pred_filter)) / 100
results['AVG_docking_scores_unfiltered'] = np.mean(
[docking_scores[i] for i in pred_filter])
if 'diversity' in criteria:
from tdc import Evaluator
evaluator = Evaluator(name='Diversity')
score = evaluator(pred_)
results['diversity'] = score
if 'validity' in criteria:
from tdc import Evaluator
evaluator = Evaluator(name='Validity')
score = evaluator(pred_)
results['validity'] = score
if 'uniqueness' in criteria:
from tdc import Evaluator
evaluator = Evaluator(name='Uniqueness')
score = evaluator(pred_)
results['uniqueness'] = score
results_all[dataset_name] = results
return results_all
if true is None:
# test set evaluation
metric_dict = bm_metric_names[self.name]
out = {}
for data_name, pred_ in pred.items():
data_name = fuzzy_search(data_name, self.dataset_names)
data_path = os.path.join(self.path, data_name)
if self.file_format == 'csv':
test = pd.read_csv(os.path.join(data_path, 'test.csv'))
elif self.file_format == 'pkl':
test = pd.read_pickle(os.path.join(data_path, 'test.pkl'))
y = test.Y.values
evaluator = eval('Evaluator(name = \'' +
metric_dict[data_name] + '\')')
out[data_name] = {
metric_dict[data_name]: round(evaluator(y, pred_), 3)
}
# If reporting accuracy across target classes
if 'target_class' in test.columns:
test['pred'] = pred_
for c in test['target_class'].unique():
data_name_subset = data_name + '_' + c
test_subset = test[test['target_class'] == c]
y_subset = test_subset.Y.values
pred_subset = test_subset.pred.values
evaluator = eval('Evaluator(name = \'' +
metric_dict[data_name_subset] + '\')')
out[data_name_subset] = {
metric_dict[data_name_subset]:
round(evaluator(y_subset, pred_subset), 3)
}
return out
else:
# validation set evaluation
if benchmark is None:
raise ValueError(
'Please specify the benchmark name for us to retrieve the standard metric!'
)
data_name = fuzzy_search(benchmark, self.dataset_names)
metric_dict = bm_metric_names[self.name]
evaluator = eval('Evaluator(name = \'' + metric_dict[data_name] +
'\')')
return {metric_dict[data_name]: round(evaluator(true, pred), 3)}
sys.path.append("..")
import os
import json
import numpy as np
import pandas as pd
import functools
from dqn import molecules
from dqn import deep_q_networks
from dqn.py.SA_Score import sascorer
from chemutil import similarity
from rdkit import Chem, DataStructs
from rdkit.Chem import AllChem, Draw, Descriptors, QED
from tdc import Oracle
qed_oracle = Oracle(name='qed')
# import matplotlib.pyplot as plt
import tensorflow as tf
from pathlib import Path
def latest_ckpt(path):
return max([
int(p.stem.split('-')[1]) for p in path.iterdir()
if p.stem[:4] == 'ckpt'
])
# basepath = '/Users/odin/sherlock_scratch/moldqn2/target_sas/mol%i_target_%.1f'
path = Path("save_qed")
seq = uniprot2seq('P49122')
# data split
from tdc.single_pred import ADME
data = ADME(name='Caco2_Wang')
split = data.get_split(method='scaffold')
from tdc.multi_pred import DTI
data = DTI(name='DAVIS')
split = data.get_split(method='cold_split', column_name='Drug')
# Molecule Generation Oracles
from tdc import Oracle
oracle = Oracle(name='GSK3B')
smiles_lst = ['CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'C[C@@H]1CCc2c(sc(NC(=O)c3ccco3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O']
oracle(smiles_lst)
oracle = Oracle(name='DRD2')
smiles_lst = ['CC(C)(C)[C@H]1CCc2c(sc(NC(=O)COc3ccc(Cl)cc3)c2C(N)=O)C1', \
'C[C@@H]1CCc2c(sc(NC(=O)c3ccco3)c2C(N)=O)C1', \
'CCNC(=O)c1ccc(NC(=O)N2CC[C@H](C)[C@H](O)C2)c(C)c1', \
'C[C@@H]1CCN(C(=O)CCCc2ccccc2)C[C@@H]1O']
oracle(smiles_lst)
oracle = Oracle(name='Hop')
oracle(['CC(=O)OC1=CC=CC=C1C(=O)O', 'C1=CC=C(C=C1)C=O'])
