def time_roundtrip(file_path: str, sample_size: int = -1):
"""Tests the amount of time it takes to encode and then decode an
entire .txt file of SMILES strings <n> times. If <sample_size> is positive,
then a random sample is taken from the file instead.
"""
curr_dir = os.path.dirname(__file__)
file_path = os.path.join(curr_dir, file_path)
# load data
with open(file_path, 'r') as file:
smiles = [line.rstrip() for line in file.readlines()]
smiles.pop(0)
if sample_size > 0:
smiles = random.sample(smiles, sample_size)
selfies = list(map(sf.encoder, smiles))
print(f"Timing {len(smiles)} SMILES from {file_path}")
# time sf.encoder
start = time.time()
for s in smiles:
sf.encoder(s)
enc_time = time.time() - start
print(f"--> selfies.encoder: {enc_time:0.7f}s")
# time sf.decoder
start = time.time()
for s in selfies:
sf.decoder(s)
dec_time = time.time() - start
print(f"--> selfies.decoder: {dec_time:0.7f}s")
def test_standardized_alphabet():
"""Tests that equivalent SMILES atom symbols are translated into the
same SELFIES atom symbol.
"""
assert sf.encoder("[C][O][N][P][F]") == "[CH0][OH0][NH0][PH0][FH0]"
assert sf.encoder("[Fe][Si]") == "[Fe][Si]"
assert sf.encoder("[Fe++][Fe+2]") == "[Fe+2][Fe+2]"
assert sf.encoder("[CH][CH1]") == "[CH1][CH1]"
def get_selfie_and_smiles_encodings_for_dataset(filename_data_set_file_smiles):
"""
Returns encoding, alphabet and length of largest molecule in SMILES and SELFIES, given a file containing SMILES molecules.
input:
csv file with molecules. Column's name must be 'smiles'.
output:
- selfies encoding
- selfies alphabet
- longest selfies string
- smiles encoding (equivalent to file content)
- smiles alphabet (character based)
- longest smiles string
"""
df = pd.read_csv(filename_data_set_file_smiles)
smiles_list = np.asanyarray(df.smiles)
smiles_alphabet=list(set(''.join(smiles_list)))
largest_smiles_len=len(max(smiles_list, key=len))
selfies_list=[]
selfies_len=[]
print('--> Translating SMILES to SELFIES...')
for individual_smile in smiles_list:
individual_selfie=selfies.encoder(individual_smile)
selfies_list.append(individual_selfie)
selfies_len.append(len(individual_selfie)-len(individual_selfie.replace('[',''))) # len of SELFIES
selfies_alphabet_pre=list(set(''.join(selfies_list)[1:-1].split('][')))
selfies_alphabet=[]
for selfies_element in selfies_alphabet_pre:
selfies_alphabet.append('['+selfies_element+']')
largest_selfies_len=max(selfies_len)
print('Finished translating SMILES to SELFIES.')
return(selfies_list, selfies_alphabet, largest_selfies_len, smiles_list, smiles_alphabet, largest_smiles_len)
def selfies(dir="../data/xyz/"):
temp = xyz_to_smiles(dir)
ret = []
for i in temp:
selfies_temp = encoder(i)
ret.append(selfies_temp)
ret = np.array(ret)
return ret
def mol2string(mol):
smiles = Chem.MolToSmiles(mol)
if string_type == 'selfies':
return encoder(smiles).split('][')
if string_type == 'deepsmiles':
string = converter.encode(smiles)
return list(string)
return list(smiles)
def mol2string(mol):
smiles = Chem.MolToSmiles(mol)
if string_type == 'SELFIES':
return encoder(smiles).split('][')
if string_type == 'DeepSMILES':
string = converter.encode(smiles)
return list(string)
return list(smiles)
def generate_ops(smi):
smi = smi.strip()
mol = Chem.MolFromSmiles(smi)
if mol is None:
print(f"None value: {smi}\n")
return []
scaffold = Chem.Scaffolds.MurckoScaffold.GetScaffoldForMol(mol)
frags = sg.get_next_murcko_fragments(scaffold)
smi_selfies = selfies.encoder(smi)
scaffold_selfies = selfies.encoder(Chem.MolToSmiles(scaffold))
data = []
data += [('SCAFFOLD', smi_selfies, scaffold_selfies),
('EXPAND', scaffold_selfies, smi_selfies)]
for frag in frags:
frag_seflies = selfies.encoder(Chem.MolToSmiles(frag))
data.append(('LOWER', scaffold_selfies, frag_seflies))
data.append(('UPPER', frag_seflies, scaffold_selfies))
return data
def __init__(self, smiles_file, percentage, vocab):
"""
smiles_file: path to the .smi file containing SMILES.
percantage: percentage of the dataset to use.
"""
super(SMILESDataset, self).__init__()
assert (0 < percentage <= 1)
self.percentage = percentage
self.vocab = vocab
# load eaqual portion of data from each tranche
self.data = self.read_smiles_file(smiles_file)
print("total number of SMILES loaded: ", len(self.data))
# convert the smiles to selfies
if self.vocab.name == "selfies":
self.data = [
sf.encoder(x) for x in self.data if sf.encoder(x) is not None
]
print("total number of valid SELFIES: ", len(self.data))
def mol2string(mol):
Chem.Kekulize(mol, clearAromaticFlags=True)
smiles = Chem.MolToSmiles(mol, canonical=False)
if string_type == 'selfies':
return encoder(smiles).split('][')
if string_type == 'deepsmiles':
string = converter.encode(smiles)
return list(string)
return list(smiles)
def create_annotations(input_file, output_file, idx2selfies, selfies2idx, max_length):
with open(input_file, 'r) as f:
for l in f:
l = l.strip().split('\t')
if len(l) > 1:
smiles = l[0]
img = l[1]
try:
selfies = [i + ']' for i in sf.encoder(l[0]).split(']')[:-1]]
selfies = [i + ']' for i in s]
except:
pass
def test_invalid_or_unsupported_smiles_encoder():
malformed_smiles = [
"",
"(",
"C(Cl)(Cl)CC[13C",
"C(CCCOC",
"C=(CCOC",
"CCCC)",
"C1CCCCC",
"C(F)(F)(F)(F)(F)F", # violates bond constraints
"C=C1=CCCCCC1", # violates bond constraints
"CC*CC", # uses wildcard
"C$C", # uses $ bond
"S[As@TB1](F)(Cl)(Br)N", # unrecognized chirality,
"SOMETHINGWRONGHERE",
"1243124124",
]
for smiles in malformed_smiles:
with pytest.raises(sf.EncoderError):
sf.encoder(smiles)
def selfies_scanner(*, parent_smiles: str):
# get the parent mol set up properly with defined aromaticity
parent_mol = Chem.MolFromSmiles(parent_smiles, sanitize=True)
Chem.rdmolops.Kekulize(parent_mol)
parent_smiles = Chem.MolToSmiles(parent_mol,
isomericSmiles=True,
kekuleSmiles=True)
logger.info(f"Generating children from: {parent_smiles}")
children = [] # finished children
spawns = 0 # counter for children produced
parent_selfies = encoder(parent_smiles)
symbols = re.findall(r"[^[]*\[([^]]*)\]",
parent_selfies) # get the SELFIES symbols into a list
for i, symb in enumerate(symbols):
if not (symb == "epsilon" or "Branch" in symb or "Ring" in symb):
if symb in ALLOWED_SUBS: # if we have rules for how to handle this symbol
for replacement in ALLOWED_SUBS[symb]:
mut_symbols = symbols.copy(
) # don't manipulate the original
mut_symbols[i] = replacement
child_symbols = [f"[{symb}]" for symb in mut_symbols]
child = "".join(child_symbols)
child_smiles = decoder(child) # get the smiles
# test that smiles is valid
try:
# same as parent, have to have explicit aromaticity
child_mol = Chem.MolFromSmiles(child_smiles,
sanitize=True)
Chem.rdmolops.Kekulize(child_mol)
child_smiles = Chem.MolToSmiles(child_mol,
isomericSmiles=True,
kekuleSmiles=True)
assert child_mol # if MolToSmiles fails, it will be a None
if child_smiles == parent_smiles: # ignore this child if it's the same as the parent
continue
except Exception: # pylint: disable=broad-except
logger.warning(
f"Produced improper SELFIES. Ignoring and trying again. Details below:"
)
logger.warning(f"Child SELFIES: {child}")
logger.warning(f"Parent SELFIES:{parent_selfies}")
logger.warning(f"Child SMILES: {child_smiles}")
logger.warning(f"Parent SMILES: {parent_smiles}")
continue
# Every good child deserves fudge
children.append(child_smiles)
spawns += 1 # update our counter
return children
def iterate_dataframe(dataset: pd.DataFrame) -> Iterable[Sentence]:
for _, row in dataset.iterrows():
res = encoder(row.smiles)
if not res:
continue
res = res.replace("]", "] ").replace(".", "DOT ")
sent = Sentence(res.strip(), use_tokenizer=plain_tokenizer)
for col, val in row.items():
if isinstance(val, float):
if val == 1.0:
sent.add_label(None, col.replace(" ", "_") + "_P ")
if val == 0.0:
sent.add_label(None, col.replace(" ", "_") + "_N ")
yield sent
def test_encoder_attribution():
smiles = "C1([O-])C=CC=C1Cl"
indices = [0, 3, 3, 3, 5, 7, 8, 10, None, None, 12]
s, am = sf.encoder(smiles, attribute=True)
# check that Cl lined up
for i, ta in enumerate(am):
if ta[1]:
assert indices[i] == ta[1][0][0], \
f'found {ta[1]}; should be {indices[i]}'
if ta[0] == '[Cl]':
for i, v in ta[1]:
if v == 'Cl':
return
raise ValueError('Failed to find Cl in attribution map')
def test_roundtrip_translation(test_path, dataset_samples):
"""Tests SMILES -> SELFIES -> SMILES translation on various datasets.
"""
# very relaxed constraints
constraints = sf.get_preset_constraints("hypervalent")
constraints.update({"P": 7, "P-1": 8, "P+1": 6, "?": 12})
sf.set_semantic_constraints(constraints)
error_path = ERROR_LOG_DIR / "{}.csv".format(test_path.stem)
with open(error_path, "w+") as error_log:
error_log.write("In, Out\n")
error_data = []
error_found = False
n_lines = sum(1 for _ in open(test_path)) - 1
n_keep = dataset_samples if (0 < dataset_samples <= n_lines) else n_lines
skip = random.sample(range(1, n_lines + 1), n_lines - n_keep)
reader = pd.read_csv(test_path, chunksize=10000, header=0, skiprows=skip)
for chunk in reader:
for in_smiles in chunk["smiles"]:
in_smiles = in_smiles.strip()
mol = Chem.MolFromSmiles(in_smiles, sanitize=True)
if (mol is None) or ("*" in in_smiles):
continue
try:
selfies = sf.encoder(in_smiles, strict=True)
out_smiles = sf.decoder(selfies)
except (sf.EncoderError, sf.DecoderError):
error_data.append((in_smiles, ""))
continue
if not is_same_mol(in_smiles, out_smiles):
error_data.append((in_smiles, out_smiles))
with open(error_path, "a") as error_log:
for entry in error_data:
error_log.write(",".join(entry) + "\n")
error_found = error_found or error_data
error_data = []
sf.set_semantic_constraints() # restore constraints
assert not error_found
def smiles2string(smiles):
if string_type == 'smiles':
string = smiles
if string_type == 'selfies':
try:
string = encoder(smiles, PrintErrorMessage=False)
except:
return None
if string_type == 'deepsmiles':
try:
string = converter.encode(smiles)
except deepsmiles.DecodeError as e:
return None
return string
def batch_mode(input_file, output_file, model_size):
outfile = open(output_file, "w")
with open(input_file, "r") as f:
for i, line in enumerate(f):
smiles_string = line.strip()
canonical_smiles = subprocess.check_output([
'java', '-cp', 'Java_dependencies/cdk-2.1.1.jar:.',
'SMILEStoCanonicalSMILES', smiles_string
])
iupac_name = translate(
selfies.encoder(
canonical_smiles.decode('utf-8').strip()).replace(
"][", "] ["), model_size)
outfile.write(
iupac_name.replace(" ", "").replace("<end>", "") + "\n")
outfile.close()
return output_file
def to_selfies(mol: Union[str, Chem.rdchem.Mol]) -> Optional[str]:
"""Convert a mol to SELFIES.
Args:
mol: a molecule or a SMILES.
Returns:
selfies: SELFIES string.
"""
if mol is None:
return None
if isinstance(mol, Chem.rdchem.Mol):
mol = to_smiles(mol)
selfies = sf.encoder(mol) # type: ignore
if selfies == -1:
return None
return selfies
def roundtrip_translation():
sf.set_semantic_constraints("hypervalent")
n_entries = 0
for chunk in make_reader():
n_entries += len(chunk)
pbar = tqdm(total=n_entries)
reader = make_reader()
error_log = open(ERROR_LOG_DIR / f"{TEST_SET_PATH.stem}.txt", "a+")
curr_idx = 0
for chunk_idx, chunk in enumerate(reader):
for in_smiles in chunk[args.col_name]:
pbar.update(1)
curr_idx += 1
if curr_idx < args.start_from:
continue
in_smiles = in_smiles.strip()
mol = Chem.MolFromSmiles(in_smiles, sanitize=True)
if (mol is None) or ("*" in in_smiles):
continue
try:
selfies = sf.encoder(in_smiles, strict=True)
out_smiles = sf.decoder(selfies)
except (sf.EncoderError, sf.DecoderError):
error_log.write(in_smiles + "\n")
tqdm.write(in_smiles)
continue
if not is_same_mol(in_smiles, out_smiles):
error_log.write(in_smiles + "\n")
tqdm.write(in_smiles)
error_log.close()
def create_tokenized_smiles_json(tokenizer, data_dir, split, config_output_name, max_length, label_filename, idx2selfies, selfies2idx):
data = {"images" : []}
start_token_id = tokenizer.encode('<start>').ids[0]
end_token_id = tokenizer.encode('<end>').ids[0]
pad_token_id = tokenizer.get_vocab_size()
print("The <start> token id is:{}\nThe <end> token id is:{}\nThe <pad> token id is {}".format(start_token_id, end_token_id, pad_token_id))
with open(os.path.join(data_dir, label_filename), "r") as f:
for i, l in enumerate(tqdm(f)):
try:
smiles, idx = l.strip().split("\t")
encoding = tokenizer.encode(smiles)
selfies = sf.encoder(smiles)
if selfies == None:
selfies = ''
selfies_cap = [selfies2idx['[start]']] + [selfies2idx[i] for i in sf.split_selfies(selfies)]
if len(selfies_cap) > max_length-1:
selfies_cap = selfies_cap[:max_length-1]
selfies_cap = selfies_cap + [selfies2idx['[end]']]
selfies_len = len(selfies_cap)
selfies_len_orig = selfies_len
while selfies_len < max_length:
selfies_cap = selfies_cap + [selfies2idx['[pad]']]
selfies_len += 1
encodingids = encoding.ids
encodingids = [start_token_id] + encodingids[:-1] #add <start> token and shorten to 150
cap_len = max_length
for j in range(0, len(encodingids)):
if encodingids[j] == pad_token_id:
cap_len = j+1
encodingids[j] = end_token_id
break
current_sample = {"filepath": data_dir, "filename": "{}".format(idx), "imgid": 0, "split": split, "sentences" : [{"tokens": encoding.tokens, "raw": smiles, "ids": encodingids , "length": cap_len, "selfies_raw": selfies, "selfies_ids": selfies_cap, "selfies_length":selfies_len_orig }] } # note if image augmentation ever happens need to introduce a sentence id token. see mscoco json for example
data["images"].append(current_sample)
except:
pass
pickle.dump(data, open(os.path.join(data_dir, config_output_name),'wb'))
del data
def __getitem__(self, idx):
# Returns tuple
# Smiles has to be in first column of the csv !!
row = self.df.iloc[idx, :]
smiles = row.smiles # needed anyway to build graph
m = Chem.MolFromSmiles(smiles)
if self.compute_selfies:
Chem.Kekulize(m)
k = Chem.MolToSmiles(m, isomericSmiles=False,
kekuleSmiles=True) # kekuleSmiles
selfie = encoder(k)
"""
if selfie != row.selfies:
print('new selfie:', selfie)
print('prev : ', row.selfies)
"""
else:
selfie = row.selfies
# 1 - Graph building
if m != None:
graph = smiles_to_nx(smiles)
else:
return None, 0, 0, 0
one_hot = {
edge: torch.tensor(self.edge_map[label])
for edge, label in (
nx.get_edge_attributes(graph, 'bond_type')).items()
}
nx.set_edge_attributes(graph, name='one_hot', values=one_hot)
try:
at_type = {
a: oh_tensor(self.at_map[label], self.num_atom_types)
for a, label in (
nx.get_node_attributes(graph, 'atomic_num')).items()
}
nx.set_node_attributes(graph, name='atomic_num', values=at_type)
except KeyError:
print('!!!! Atom type to one-hot error for input ', smiles,
' ignored')
return None, 0, 0, 0
at_charge = {
a: oh_tensor(self.charges_map[label], self.num_charges)
for a, label in (
nx.get_node_attributes(graph, 'formal_charge')).items()
}
nx.set_node_attributes(graph, name='formal_charge', values=at_charge)
try:
hydrogens = {
a: torch.tensor(self.chi_map[label], dtype=torch.float)
for a, label in (
nx.get_node_attributes(graph, 'num_explicit_hs')).items()
}
nx.set_node_attributes(graph,
name='num_explicit_hs',
values=hydrogens)
except KeyError:
print(
'!!!! Number of explicit hydrogens to one-hot error for input ',
smiles, ' ignored')
return None, 0, 0, 0
aromatic = {
a: torch.tensor(self.chi_map[label], dtype=torch.float)
for a, label in (
nx.get_node_attributes(graph, 'is_aromatic')).items()
}
nx.set_node_attributes(graph, name='is_aromatic', values=aromatic)
at_chir = {
a: torch.tensor(self.chi_map[label], dtype=torch.float)
for a, label in (
nx.get_node_attributes(graph, 'chiral_tag')).items()
}
nx.set_node_attributes(graph, name='chiral_tag', values=at_chir)
# to dgl
g_dgl = dgl.DGLGraph()
node_features = [
'atomic_num', 'formal_charge', 'num_explicit_hs', 'is_aromatic',
'chiral_tag'
]
g_dgl.from_networkx(nx_graph=graph,
node_attrs=node_features,
edge_attrs=['one_hot'])
N = g_dgl.number_of_nodes()
g_dgl.ndata['h'] = torch.cat(
[g_dgl.ndata[f].view(N, -1) for f in node_features], dim=1)
if self.graph_only: # give only the graph (to encode in latent space)
return g_dgl, 0, 0, 0
# 2 - Smiles / selfies to integer indices array
if self.language == 'selfies':
a, valid_flag = self.selfies_to_hot(selfie)
if valid_flag == 0: # no one hot encoding for this selfie, ignore
print('!!! Selfie to one-hot failed with current alphabet')
return None, 0, 0, 0
else:
a = np.zeros(self.max_len)
idces = [self.char_to_index[c] for c in smiles]
a[:len(idces)] = idces
# 3 - Optional props and affinities
props, targets = 0, 0
if len(self.props) > 0:
props = np.array(row[self.props], dtype=np.float32)
if len(self.targets) > 0 and self.binned_scores:
targets = np.array(row[self.targets],
dtype=np.int64) # for torch.long class labels
elif len(self.targets) > 0:
targets = np.array(row[self.targets],
dtype=np.float32) # for torch.float values
targets = np.nan_to_num(targets) # if nan somewhere, change to 0.
return g_dgl, a, props, targets
def test_selfies_split(self) -> None:
"""Test tokenization by selfies package has not changed."""
benzene = 'c1ccccc1'
encoded_selfies = sf.encoder(benzene)
# '[c][c][c][c][c][c][Ring1][Branch1_1]' v0.2.4
# '[C][=C][C][=C][C][=C][Ring1][Branch1_2]' v1.0.2 (no aromatic)
# sf.split_selfies returns generator
symbols_benzene = list(sf.split_selfies(encoded_selfies))
# before selfies 2.0.0 the last token is [Branch1_2]
self.assertListEqual(
symbols_benzene,
['[C]', '[=C]', '[C]', '[=C]', '[C]', '[=C]', '[Ring1]', '[=Branch1]'],
)
for smiles, ground_truth in [
(
'c1cnoc1',
# before selfies 2.0.0 the 2 last token are [Expl=Ring1] and [Branch1_1]
['[C]', '[C]', '[=N]', '[O]', '[C]', '[=Ring1]', '[Branch1]'],
),
(
'[O-][n+]1ccccc1S',
# before selfies 2.0.0 it is: [O-expl], [N+expl] and [=Branch1_2]
[
'[O-1]',
'[N+1]',
'[=C]',
'[C]',
'[=C]',
'[C]',
'[=C]',
'[Ring1]',
'[=Branch1]',
'[S]',
],
),
(
'c1snnc1-c1ccccn1',
# before selfies 2.0.0 it is: [Expl=Ring1], [Branch1_1] and [Branch1_2]
[
'[C]',
'[S]',
'[N]',
'[=N]',
'[C]',
'[=Ring1]',
'[Branch1]',
'[C]',
'[=C]',
'[C]',
'[=C]',
'[C]',
'[=N]',
'[Ring1]',
'[=Branch1]',
],
),
]:
self.assertListEqual(
# list wrapping version
split_selfies(sf.encoder(smiles)),
ground_truth,
)
'''
Written by Jan H. Jensen 2019
'''
from selfies import encoder, decoder
import pandas as pd
from rdkit import Chem
pd.set_option('max_colwidth', 200)
df = pd.read_csv('ZINC_250k.smi', sep=" ", header=None)
df.columns = ["smiles"]
rows = 1000
symbols_list = []
for index, row in df.iterrows():
smiles = row['smiles']
mol = Chem.MolFromSmiles(smiles)
Chem.Kekulize(mol, clearAromaticFlags=True)
smiles = Chem.MolToSmiles(mol)
symbols = encoder(smiles).split('][')
for symbol in symbols:
symbol = symbol.replace(']', '').replace('[', '')
if symbol not in symbols_list:
symbols_list.append(symbol)
print(symbols_list)
def selfies_substitution(*,
parent_smiles: str,
n_children: int = 100,
mut_rate: float = 0.03,
mut_min: int = 1,
mut_max: int = 2,
max_trials: int = 100
):
"""
This function takes a parent molecule, and generates a number of children derived from it, via converting
the parent into SELFIES format and substituting symbols for other symbols, based on primitive chemical rules.
:param parent_smiles: The smiles of the parent molecule.
:param n_children: How many children to produce.
:param mut_rate: How frequent should mutations be? 0.0 to 1.0
:param mut_min: What is the min number of mutations to allow in a given child, relative to the parent?
:param mut_max: Same as above but the max.
:param max_trials: number of attempts to create valid mutations before moving on, useful for pathological selfies
:return:
"""
# get the parent mol set up properly with defined aromaticity
parent_mol = Chem.MolFromSmiles(parent_smiles, sanitize=True)
Chem.rdmolops.Kekulize(parent_mol)
parent_smiles = Chem.MolToSmiles(parent_mol, isomericSmiles=True, kekuleSmiles=True)
logger.info(f"Generating children from: {parent_smiles}")
children = [] # finished children
spawns = 0 # counter for children produced
parent_selfies = encoder(parent_smiles)
symbols = re.findall(r"[^[]*\[([^]]*)\]", parent_selfies) # get the SELFIES symbols into a list
while spawns < n_children: # try to produce the correct number of children
muts = 0
mutations = [] # which parts of the SELFIES to remove
mut_symbols = symbols.copy() # don't manipulate the original
mut_positions = list(range(len(symbols))) # need the index
t = 0
while (muts < mut_min) and (t <= max_trials):
random.shuffle(mut_positions) # shuffle the order so that mutations will be random
for pos in mut_positions: # try to mutate
if pos not in mutations:
if random.random() <= mut_rate:
# ignore special symbols, leave them alone
if not (symbols[pos] == "epsilon" or "Branch" in symbols[pos] or "Ring" in symbols[pos]):
if symbols[pos] in ALLOWED_SUBS: # if we have rules for how to handle this symbol
mutations.append(pos) # record which symbol this is
muts += 1 # record the intention to mutate
if muts == mut_max:
# when we're done, stop looking
break
t += 1
if t > max_trials:
logger.warning(f"Failed to produce any selfies after {max_trials} trials. Returning empty list.")
return list()
# for each planned mutation, actually execute it, on the non-shuffled original SELFIES list
for index in sorted(mutations, reverse=True):
mut_symbols[index] = random.choice(ALLOWED_SUBS[mut_symbols[index]])
# convert the new child into a SELFIES (rather than a list)
child_symbols = [f"[{symb}]" for symb in mut_symbols]
child = "".join(child_symbols)
child_smiles = decoder(child) # get the smiles
# test that smiles is valid
try:
# same as parent, have to have explicit aromaticity
child_mol = Chem.MolFromSmiles(child_smiles, sanitize=True)
Chem.rdmolops.Kekulize(child_mol)
child_smiles = Chem.MolToSmiles(child_mol, isomericSmiles=True, kekuleSmiles=True)
assert child_mol # if MolToSmiles fails, it will be a None
if child_smiles == parent_smiles: # ignore this child if it's the same as the parent
continue
except Exception: # pylint: disable=broad-except
logger.warning(f"Produced improper SELFIES. Ignoring and trying again. Details below:")
logger.warning(f"Child SELFIES: {child}")
logger.warning(f"Parent SELFIES:{parent_selfies}")
logger.warning(f"Child SMILES: {child_smiles}")
logger.warning(f"Parent SMILES: {parent_smiles}")
continue
# Every good child deserves fudge
children.append(child_smiles)
spawns += 1 # update our counter
return children
num_iterations = 1
results_dir = du.make_clean_results_dir()
total_time = time.time()
for i in range(num_iterations):
for beta in beta_params:
max_fitness_collector = []
image_dir, saved_models_dir, data_dir = du.make_clean_directories(beta, results_dir, i)
torch.cuda.empty_cache()
writer = SummaryWriter()
smiles_all_counter = train( num_generations = 100,
generation_size = 500,
starting_selfies = [encoder('C')],
max_molecules_len = 81,
disc_epochs_per_generation = 10,
disc_enc_type = 'properties_rdkit',
disc_layers = [100, 10],
training_start_gen = 0,
device = 'cuda',
properties_calc_ls = ['logP', 'SAS', 'RingP', 'QED'],
num_processors = multiprocessing.cpu_count(),
beta = beta,
max_fitness_collector = max_fitness_collector,
impose_time_adapted_pen = True
)
print('Total time: ', (time.time()-total_time)/60, ' mins')
'VAE_dependencies/Saved_models/VAE_decode_epoch_{}'.format(
settings['training_VAE']['num_epochs']))
#plot epoch vs reconstruction loss / quality
print(recons_quality_valid, recons_quality_train, recons_loss)
if settings['plot']['plot_quality']:
line1, = plt.plot(recons_quality_valid, label='Validation set')
line2, = plt.plot(recons_quality_train, label='Training set')
plt.xlabel('Epochs')
plt.ylabel('Reconstruction Quality (%)')
plt.legend(handles=[line1, line2])
plt.show()
if settings['plot']['plot_loss']:
plt.plot(recons_loss)
plt.xlabel('Epochs')
plt.ylabel('Reconstruction Loss')
plt.show()
else:
print('Linear interpolation of 10 steps between ' + test_mol1 +
' and ' + test_mol2 + ':')
mol_inter = linear_interpolation(encoder(test_mol1),
encoder(test_mol2), 10)
print(mol_inter)
with open('COMPLETED', 'w') as content:
content.write('exit code: 0')
except AttributeError:
_, error_message, _ = sys.exc_info()
print(error_message)
print(error_message)
def test_roundtrip_translation(test_name, column_name, dataset_samples):
"""Tests a roundtrip SMILES -> SELFIES -> SMILES translation of the
SMILES examples in QM9, NonFullerene, Zinc, etc.
"""
# modify semantic bond constraints
sf.set_semantic_constraints({
'H': 1, 'F': 1, 'Cl': 1, 'Br': 1, 'I': 1,
'O': 2, 'O+1': 3, 'O-1': 1,
'N': 6, 'N+1': 4, 'N-1': 2,
'C': 4, 'C+1': 5, 'C-1': 3,
'S': 6, 'S+1': 7, 'S-1': 5,
'P': 7, 'P+1': 8, 'P-1': 6,
'?': 8,
})
# file I/O
curr_dir = os.path.dirname(__file__)
test_path = os.path.join(curr_dir, 'test_sets', test_name + ".txt")
error_path = os.path.join(curr_dir,
'error_sets',
"errors_{}.csv".format(test_name))
# create error directory
os.makedirs(os.path.dirname(error_path), exist_ok=True)
error_list = []
# add header in error log text file
with open(error_path, "w+") as error_log:
error_log.write("In, Out\n")
error_found_flag = False
# make pandas reader
N = sum(1 for _ in open(test_path)) - 1
S = dataset_samples if (0 < dataset_samples <= N) else N
skip = sorted(random.sample(range(1, N + 1), N - S))
reader = pd.read_csv(test_path,
chunksize=10000,
header=0,
skiprows=skip)
# roundtrip testing
for chunk in reader:
for in_smiles in chunk[column_name]:
# check if SMILES in chunk is a valid RDKit molecule.
# if not, skip testing
# All inputted SMILES must be valid
# RDKit Mol objects to be encoded.
if (MolFromSmiles(in_smiles) is None) or ('*' in in_smiles):
continue
# encode SELFIE string
selfies = sf.encoder(in_smiles)
# if unable to encode SMILES, write to list of errors
if selfies is None:
error_list.append((in_smiles, ''))
continue
# take encoeded SELFIES and decode
out_smiles = sf.decoder(selfies)
# compare original SMILES to decoded SELFIE string.
# if not the same string, write to list of errors.
if not is_same_mol(in_smiles, out_smiles):
error_list.append((in_smiles, str(out_smiles)))
# open and write all errors to errors_{test_name}.csv
with open(error_path, "a") as error_log:
for error in error_list:
error_log.write(','.join(error) + "\n")
error_found_flag = error_found_flag or error_list
error_list = []
sf.set_semantic_constraints() # restore defaults
assert not error_found_flag
def test_malformed_smiles_encoder():
"""Tests selfies.encoder() terminates on a malformed SMILES."""
sf.encoder("C(Cl)(Cl)CC[13C")
assert True
def test_roundtrip_translation():
"""Tests a roundtrip SMILES -> SELFIES -> SMILES translation of the
SMILES examples in QM9, NonFullerene, Zinc, etc.
"""
# modify constraints
constraints = sf.get_semantic_constraints()
constraints['N'] = 6
constraints['Br'] = 7
constraints['Cl'] = 7
constraints['I'] = 7
sf.set_semantic_constraints(constraints)
# file I/O
ckpt_path = os.path.join(curr_dir, 'checkpoints', 'emolecule_ckpt.txt')
error_path = os.path.join(curr_dir, 'error_sets', 'errors_emolecules.csv')
# check if a previous checkpoint exists to continue tests
if os.path.exists(ckpt_path):
with open(ckpt_path, 'r') as ckpt_file:
checkpoint = int(ckpt_file.readlines()[0])
# if no path to a checkpoint exists,
# create a new directory for error logging and checkpoints
else:
os.makedirs(os.path.dirname(ckpt_path), exist_ok=True)
os.makedirs(os.path.dirname(error_path), exist_ok=True)
with open(error_path, "w+") as error_log:
error_log.write("In, Out\n")
checkpoint = -1
error_list = []
error_found_flag = False
# make pandas reader
reader = pd.read_csv(EMOL_PATH,
chunksize=10000,
compression='gzip',
delimiter=' ',
header=0)
# roundtrip testing
for chunk_idx, chunk in enumerate(reader):
if chunk_idx <= checkpoint:
continue
for in_smiles in chunk[COL_NAME]:
# check if SMILES in chunk is a valid RDKit molecule.
# if not, skip testing
# All inputted SMILES must be valid
# RDKit Mol objects to be encoded.
if (MolFromSmiles(in_smiles) is None) or ('*' in in_smiles):
continue
# encode selfies
selfies = sf.encoder(in_smiles)
# if unable to encode SMILES, write to list of errors
if selfies is None:
error_list.append((in_smiles, ''))
continue
# take encoeded SELFIES and decode
out_smiles = sf.decoder(selfies)
# compare original SMILES to decoded SELFIE string.
# if not the same string, write to list of errors.
if not is_same_mol(in_smiles, out_smiles):
error_list.append((in_smiles, out_smiles))
# open and write all errors to errors_emolecule.csv
with open(error_path, "a") as error_log:
for error in error_list:
error_log.write(','.join(error) + "\n")
error_found_flag = error_found_flag or error_list
error_list = []
# create checkpoint from the current pandas reader chunk,
# to load from and continue testing.
with open(ckpt_path, 'w+') as ckpt_file:
ckpt_file.write(str(chunk_idx))
sf.set_semantic_constraints() # restore defaults
os.remove(ckpt_path) # remove checkpoint
assert not error_found_flag
def main(args):
with open(args.input_file, 'r') as f:
with open(args.output_file, 'w') as w:
for l in f:
l = l.strip()
w.write('{}\n'.format(sf.encoder(l)))
