def testRingEncoding(self):
ringconverter = ds.Converter(rings=True)
allconverter = ds.Converter(rings=True, branches=True)
data = [ # smi, DeepSmiles/RC, DeepSMILES/RC+PN (if different)
("C1CCCC1", "CCCCC5"),
("C%10CCC%10", "CCCC4"),
("C1CCCCCCCCC1", "CCCCCCCCCC%10"),
("C1CC(OC)CC1", "CCC(OC)CC5", "CCCOC))CC5"),
(r"N1CC=C/1\Br", r"NCC=C/4\Br"),
(r"N\1CC=C1\Br", r"NCC=C/4\Br"),
(r"C1=C/CCCCCC/1", "C=C/CCCCCC/8"),
(r"C\1=C/CCCCCC1", "C=C/CCCCCC/8"),
(r"C\1=C/CCCCCC/1", "C=C/CCCCCC/8"),
("C1N[C@@]12CO2", "CN[C@@]3CO3"),
("[C@@]12(NC1)CO2", "[C@@](NC3)CO3", "[C@@]NC3))CO3"),
("CC1CCCO[C@]21CCCCO2", "CCCCCO[C@@]6CCCCO6"),
("CC1CCCO[C@@]12CCCCO2", "CCCCCO[C@@]6CCCCO6"),
("NC[C@]12CCCC1C3CC2CC3", "NC[C@]CCCC5CCC8CC5"),
("NC[C@]12CCCC2C3CC1CC3", "NC[C@@]CCCC5CCC8CC5"),
("C2C1=C/CCCCCC/12", "CC=C/CCCCCC/89"),
("C1C2=C/CCCCCC1/2", "CC=C/CCCCCC9/8"),
]
for d in data:
encoded = ringconverter.encode(d[0])
self.assertEqual(d[1], encoded)
encodedB = allconverter.encode(d[0])
if len(d) == 2:
self.assertEqual(d[1], encodedB)
else:
self.assertEqual(d[2], encodedB)
def testRingDecoding(self):
ringconverter = ds.Converter(rings=True)
allconverter = ds.Converter(rings=True, branches=True)
data = [ # smi, DeepSmiles/RC
("C1CCC1", "CCCC4"),
("C2CC1CCC1C2", "CCCCCC4C7"),
("c1c[nH]cc1", "cc[nH]cc5"),
("C1CCCCCCCCC1", "CCCCCCCCCC%10"),
("C1CCCCCCCCC1", "CCCCCCCCCC%(10)"),
("CCCCCCC1CCC1", "CCCCCCCCCC%(4)"),
("C1=C/CCCCCC/1", "C=C/CCCCCC/8"),
("C1C1C2C2C3C3C4C4C5C5C6C6C7C7C8C8C9C9C%10C%10",
"CC2CC2CC2CC2CC2CC2CC2CC2CC2CC2"),
("C(CS)N", "C(CS)N", ""),
("C[C@@H]1CCCO[C@]12CCCCO2", "C[C@@H]CCCO[C@]6CCCCO6"),
("C2C1=C/CCCCCC/12", "CC=C/CCCCCC/89"),
("C1C2=C/CCCCCC1/2", "CC=C/CCCCCC9/8"),
]
for d in data:
decoded = ringconverter.decode(d[1])
self.assertEqual(d[0], decoded)
minput = d[2] if len(d) == 3 else d[1]
if minput:
decodedB = allconverter.decode(minput)
self.assertEqual(d[0], decodedB)
def testStringRep(self):
self.assertEqual(str(ds.Converter()),
"Converter(rings=False, branches=False)")
self.assertEqual(str(ds.Converter(rings=True)),
"Converter(rings=True, branches=False)")
self.assertEqual(str(ds.Converter(branches=True)),
"Converter(rings=False, branches=True)")
self.assertEqual(str(ds.Converter(rings=True, branches=True)),
"Converter(rings=True, branches=True)")
def testDecodingExceptions(self):
converter = ds.Converter(rings=True, branches=True)
data = [
"C8", "C))I", "%10C", "9C", "CCCCCC%(3CC", "C%(100)", "C[C@@CCl",
"C%CC", "-5cc[nH]9"
]
for dsmi in data:
self.assertRaises(ds.DecodeError, converter.decode, dsmi)
# Test just the ring decoder
converter = ds.Converter(rings=True)
data = ["C8", "%10C", "9C", "C%(100)", "C[C@@CCl", "C%CC", "-5cc[nH]9"]
for dsmi in data:
self.assertRaises(ds.DecodeError, converter.decode, dsmi)
def main():
converter = deepsmiles.Converter(rings=True, branches=True)
client = MongoClient("localhost:27018")
db = client['COCONUT2020-07']
collection = db.uniqueNaturalProduct.aggregate([{'$project': {'_id': 0, 'coconut_id': 1, 'unique_smiles': 1}}])
allnp = pd.DataFrame(list(collection))
for index, row in allnp.iterrows():
coconut_id = row['coconut_id']
smiles = row["unique_smiles"]
deep_smiles = converter.encode(smiles)
db.uniqueNaturalProduct.update_one({'coconut_id': coconut_id}, {"$set": {"deep_smiles": deep_smiles}})
# for each row - get unique_smiles -> convert to deep smiles and save at deep_smiles
# db.uniqueNaturalProduct.update_one({'coconut_id': data[0]}, {"$set": {"name": data[2]}})
print("done")
def testRoundTripRingClosures(self):
smi = "C%(1)C%(2)C%(3)C%(4)C%(5)C%(6)C%(7)C%(8)C%(9)C%(10)C%(11)C%(12)C%(13)C%(14)C%(15)C%(16)C%(17)C%(18)C%(19)C%(20)C%(21)C%(22)C%(23)C%(24)C%(25)C%(26)C%(27)C%(28)C%(29)C%(30)C%(31)C%(32)C%(33)C%(34)C%(35)C%(36)C%(37)C%(38)C%(39)C%(40)C%(41)C%(42)C%(43)C%(44)C%(45)C%(46)C%(47)C%(48)C%(49)C%(50)C%(51)C%(52)C%(53)C%(54)C%(55)C%(56)C%(57)C%(58)C%(59)C%(60)C%(61)C%(62)C%(63)C%(64)C%(65)C%(66)C%(67)C%(68)C%(69)C%(70)C%(71)C%(72)C%(73)C%(74)C%(75)C%(76)C%(77)C%(78)C%(79)C%(80)C%(81)C%(82)C%(83)C%(84)C%(85)C%(86)C%(87)C%(88)C%(89)C%(90)C%(91)C%(92)C%(93)C%(94)C%(95)C%(96)C%(97)C%(98)C%(99)C%(100)C%(100)C%(99)C%(98)C%(97)C%(96)C%(95)C%(94)C%(93)C%(92)C%(91)C%(90)C%(89)C%(88)C%(87)C%(86)C%(85)C%(84)C%(83)C%(82)C%(81)C%(80)C%(79)C%(78)C%(77)C%(76)C%(75)C%(74)C%(73)C%(72)C%(71)C%(70)C%(69)C%(68)C%(67)C%(66)C%(65)C%(64)C%(63)C%(62)C%(61)C%(60)C%(59)C%(58)C%(57)C%(56)C%(55)C%(54)C%(53)C%(52)C%(51)C%(50)C%(49)C%(48)C%(47)C%(46)C%(45)C%(44)C%(43)C%(42)C%(41)C%(40)C%(39)C%(38)C%(37)C%(36)C%(35)C%(34)C%(33)C%(32)C%(31)C%(30)C%(29)C%(28)C%(27)C%(26)C%(25)C%(24)C%(23)C%(22)C%(21)C%(20)C%(19)C%(18)C%(17)C%(16)C%(15)C%(14)C%(13)C%(12)C%(11)C%(10)C%(9)C%(8)C%(7)C%(6)C%(5)C%(4)C%(3)C%(2)C%(1)"
for branches in [True, False]:
converter = ds.Converter(rings=True, branches=branches)
encoded = converter.encode(smi)
decoded = converter.decode(encoded)
self.assertTrue("%(100)" in decoded)
def testBranchEncoding(self):
branchconverter = ds.Converter(branches=True)
allconverter = ds.Converter(rings=True, branches=True)
data = [ # smi, DeepSmiles/PN, DeepSMILES/RC+PN (if different)
("C(O)C", "CO)C"), ("C([O])C", "C[O])C"), ("C(OF)C", "COF))C"),
("C(F)(F)C", "CF)F)C"), ("C(Cl)(Cl)C", "CCl)Cl)C"),
("C(=O)Cl", "C=O)Cl"), ("C(OC(=O)Cl)I", "COC=O)Cl)))I"),
("[C@@H](Br)(Cl)I", "[C@@H]Br)Cl)I"),
("B(c1ccccc1)(O)O", "Bc1ccccc1))))))O)O", "Bcccccc6))))))O)O"),
("Cn1cccc-2nccc12", "Cn1cccc-2nccc12", "Cnccccnccc9-5")
]
for d in data:
encoded = branchconverter.encode(d[0])
self.assertEqual(d[1], encoded)
encodedB = allconverter.encode(d[0])
if len(d) == 2:
self.assertEqual(d[1], encodedB)
else:
self.assertEqual(d[2], encodedB)
def sm2ds(line):
# Takes schwaller's preprocessed SMILES and turns them into deepSMILES
converter = ds.Converter(rings=True, branches=True)
line = line.replace(" ", "")
molecules = line.split(".")
new_line = []
for molecule in molecules:
new_molecule = converter.encode(molecule)
new_line.append(new_molecule)
new_line = ".".join(new_line)
return new_line
def ds2sm(line):
# Takes a deepSMILES line and turns it into SMILES
converter = ds.Converter(rings=True, branches=True)
line = line.replace(" ", "")
molecules = line.split(".")
new_line = []
for molecule in molecules:
new_molecule = converter.decode(molecule)
new_line.append(new_molecule)
new_line = ".".join(new_line)
return new_line
def testBranchDecoding(self):
branchconverter = ds.Converter(branches=True)
data = [ # smi, DeepSmiles/PN
("COC", "COC"),
("C(O)C", "CO)C"),
("C(=O)C", "C=O)C"),
("C[O]C", "C[O]C"),
("C(OC(=O)Cl)I", "COC=O)Cl)))I"),
("C(F)(F)C", "CF)F)C"),
("Cn1ccnc1", "Cn1ccnc1"),
("c1ccn(cc1)O", "c1ccncc1))O"),
("Cn1cccc-2nccc12", "Cn1cccc-2nccc12"),
]
for d in data:
decoded = branchconverter.decode(d[1])
self.assertEqual(d[0], decoded)
def convert_smiles(smiles=False, deep=False):
'''
smiles and deep must be str format
Converts from SMILES to DeepSMILES and vice versa.
Which ever has a string provided, will convert to the other.
If strings are proivded for both, then nothing happens
'''
converter = deepsmiles.Converter(rings=True, branches=True)
if smiles and deep:
print('Only provide a string for one of smiles or deep')
return()
if smiles: # Convert from SMILES to DeepSMILES
deep_string = converter.encode(smiles)
return deep_string
if deep: # Convert from DeepSMILES to SMILES
smiles_string = converter.decode(deep)
return smiles_string
def main():
parser = argparse.ArgumentParser(description="Predicting test images")
# Input Arguments
parser.add_argument('--input',
help='Enter the input filename',
required=True)
args = parser.parse_args()
tokenizer = pickle.load(open("tokenizer.pkl", "rb"))
# Prediction model parameters
embedding_dim = 600
units = 1024
vocab_size = len(tokenizer.word_index) + 1
optimizer = tf.keras.optimizers.Adam(learning_rate=0.0005)
encoder = I2S_Model.CNN_Enc(embedding_dim)
decoder = I2S_Model.RNN_Dec(embedding_dim, units, vocab_size)
# Initialize DeepSMILES
converter = deepsmiles.Converter(rings=True, branches=True)
with open('Predictions.txt', 'w') as f:
print(datetime.now().strftime('%Y/%m/%d %H:%M:%S'),
"Predictions\n\n",
flush=True,
file=f)
result = predictor(args.input,
encoder=encoder,
decoder=decoder,
optimizer=optimizer,
tokenizer=tokenizer)
print(
converter.decode(''.join(result).replace("<start>",
"").replace("<end>", "")),
'\tPredictedSmiles',
flush=True,
file=f,
)
def convert_to_deepsmile(smiles_list):
"""
Convert smiles list to deepsmiles list. Make sure the smiles list is canonical.
Args:
smiles_list:
Returns: deepsmiles list
"""
converter = deepsmiles.Converter(rings=True, branches=True)
deep_lst = [converter.encode(smi) for smi in smiles_list]
decoded_lst = []
final_deep_lst = []
num_decode, num_recover = 0, 0
for i in range(len(deep_lst)):
try:
decoded = converter.decode(deep_lst[i])
except deepsmiles.DecodeError as e:
decoded = None
print("DecodeError! Error message was {}".format(e.message))
decoded_lst.append(decoded)
if decoded:
num_decode += 1
decoded_can_lst = []
for item in decoded_lst:
if Chem.MolFromSmiles(item) is not None:
can_item = Chem.MolToSmiles(Chem.MolFromSmiles(item))
decoded_can_lst.append(can_item)
else:
decoded_can_lst.append(None)
for i in range(len(smiles_list)):
if smiles_list[i] == decoded_can_lst[i]:
final_deep_lst.append(deep_lst[i])
return final_deep_lst
import os, sys
sys.path.append('VAE_dependencies')
sys.path.append('../../VAE_dependencies')
print(sys.argv)
import numpy as np
import yaml
import torch
from torch import nn
from random import shuffle
from data_loader import multiple_smile_to_hot, grammar_one_hot_to_smile
import pandas as pd
from GPlus2S import GrammarPlusToSMILES, IncludeRingsForSMILES
import deepsmiles
converter = deepsmiles.Converter(
rings=True, branches=True) # Coverter object, described by authors
from rdkit.Chem import MolFromSmiles
from rdkit import rdBase
rdBase.DisableLog('rdApp.error')
import time
def _make_dir(directory):
os.makedirs(directory)
def save_models(encoder, decoder, epoch):
out_dir = './saved_models/{}'.format(epoch)
_make_dir(out_dir)
def decode(generated, start='<M>', end='</M>'):
generated = DeepSMILESLanguageModelUtils.extract(generated, start, end)
converter = deepsmiles.Converter(rings=True, branches=True)
return converter.decode(generated)
def deepsml(x):
ldeep={}
for i,m in (x.items()):
converter = deepsmiles.Converter(rings=True, branches=True)
ldeep[i]=converter.encode(m)
return ldeep
def testDecodingBasic(self):
converter = ds.Converter()
dsmi = converter.decode("C")
self.assertEqual("C", dsmi)
from rdkit import Chem
from rdkit.Chem import AllChem
import re
import deepsmiles
f = open('allspec_90')
ccc = 0
for line in f:
if re.search('pred', line):
try:
smi = deepsmiles.Converter(rings=True, branches=True).decode(
line.split()[1].split('<')[0].strip())
except:
smi = None
if smi:
try:
csmi = Chem.MolToSmiles(Chem.MolFromSmiles(smi))
except:
pass
if csmi:
l = []
mm = Chem.MolFromSmiles(csmi)
mm = AllChem.AddHs(mm)
for at in mm.GetAtoms():
l.append(at.GetSymbol())
cs = l.count('C')
hs = l.count('H')
os = l.count('O')
print('pred', csmi, cs, hs, os)
if re.search('real', line):
try:
tokenizer = pickle.load(open("tokenizer.pkl","rb"))
#Prediction model parameters
embedding_dim = 600
units = 1024
vocab_size = len(tokenizer.word_index) + 1
optimizer = tf.keras.optimizers.Adam(learning_rate=0.0005)
encoder = I2S_Model.CNN_Encoder(embedding_dim)
decoder = I2S_Model.RNN_Decoder(embedding_dim, units, vocab_size)
#Initialize DeepSMILES
converter = deepsmiles.Converter(rings=True, branches=True)
#Evaluator
def evaluate(image):
maxlength = 74 #should be determined from running the training script
hidden = decoder.reset_state(batch_size=1)
temp_input = tf.expand_dims(I2S_evalData.load_image(image)[0], 0)
img_tensor_val = I2S_evalData.image_features_extract_model(temp_input)
img_tensor_val = tf.reshape(img_tensor_val, (img_tensor_val.shape[0], -1, img_tensor_val.shape[3]))
features = encoder(img_tensor_val)
dec_input = tf.expand_dims([tokenizer.word_index['<start>']], 0)
"""
if not mol:
return None
if random_type == "unrestricted":
return rkc.MolToSmiles(mol, canonical=False, doRandom=True, isomericSmiles=False)
if random_type == "restricted":
new_atom_order = list(range(mol.GetNumAtoms()))
random.shuffle(new_atom_order)
random_mol = rkc.RenumberAtoms(mol, newOrder=new_atom_order)
return rkc.MolToSmiles(random_mol, canonical=False, isomericSmiles=False)
raise ValueError("Type '{}' is not valid".format(random_type))
DEEPSMI_CONVERTERS = {
"rings": ds.Converter(rings=True),
"branches": ds.Converter(branches=True),
"both": ds.Converter(rings=True, branches=True)
}
def to_deepsmiles(smi, converter="both"):
"""
Converts a SMILES strings to the DeepSMILES alternative.
:param smi: SMILES string.
:return : A DeepSMILES string.
"""
return DEEPSMI_CONVERTERS[converter].encode(smi)
def from_deepsmiles(deepsmi, converter="both"):
output = open(args.output_fn, 'w')
rings = args.rings
branches = args.branches
do_encode = args.do_encode
do_decode = args.do_decode
if do_decode:
do_encode = False
assert(not (do_encode and do_decode))
if not (rings or branches):
print("use at least --no-rings or --no-branches",
file=sys.stderr)
sys.exit(1)
count = 0
# work ----------------------------------------------
smi_supplier = RobustSmilesSupplier(input_fn)
converter = deepsmiles.Converter(rings, branches)
if do_encode:
for smi, name in smi_supplier:
deep_smi = encode(converter, smi)
print("%s\t%s" % (deep_smi, name), file=output)
count += 1
else: # decode
for deep_smi, name in smi_supplier:
smi = decode(converter, deep_smi)
if smi != None:
print("%s\t%s" % (smi, name), file=output)
count += 1
after = time.time()
dt = after - before
print("%d molecules at %.2f mol/s" % (count, count / dt), file=sys.stderr)
