liuganghuggingface
commited on
Upload graph_decoder/molecule_utils.py with huggingface_hub
Browse files- graph_decoder/molecule_utils.py +355 -0
graph_decoder/molecule_utils.py
ADDED
@@ -0,0 +1,355 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Copyright 2024 the Llamole team.
|
2 |
+
#
|
3 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
4 |
+
# you may not use this file except in compliance with the License.
|
5 |
+
# You may obtain a copy of the License at
|
6 |
+
#
|
7 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
8 |
+
#
|
9 |
+
# Unless required by applicable law or agreed to in writing, software
|
10 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
11 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
12 |
+
# See the License for the specific language governing permissions and
|
13 |
+
# limitations under the License.
|
14 |
+
|
15 |
+
from rdkit import Chem, RDLogger
|
16 |
+
|
17 |
+
RDLogger.DisableLog("rdApp.*")
|
18 |
+
|
19 |
+
import re
|
20 |
+
import random
|
21 |
+
import logging
|
22 |
+
from rdkit import Chem
|
23 |
+
from typing import List, Tuple, Optional
|
24 |
+
random.seed(0)
|
25 |
+
import torch
|
26 |
+
|
27 |
+
bond_dict = [
|
28 |
+
None,
|
29 |
+
Chem.rdchem.BondType.SINGLE,
|
30 |
+
Chem.rdchem.BondType.DOUBLE,
|
31 |
+
Chem.rdchem.BondType.TRIPLE,
|
32 |
+
Chem.rdchem.BondType.AROMATIC,
|
33 |
+
]
|
34 |
+
|
35 |
+
ATOM_VALENCY = {6: 4, 7: 3, 8: 2, 9: 1, 15: 3, 16: 2, 17: 1, 35: 1, 53: 1}
|
36 |
+
|
37 |
+
logger = logging.getLogger(__name__)
|
38 |
+
|
39 |
+
def check_polymer(smiles):
|
40 |
+
if "*" in smiles:
|
41 |
+
monomer = smiles.replace("*", "[H]")
|
42 |
+
if mol2smiles(get_mol(monomer)) is None:
|
43 |
+
logger.warning(f"Invalid polymerization point")
|
44 |
+
return False
|
45 |
+
else:
|
46 |
+
return True
|
47 |
+
return True
|
48 |
+
|
49 |
+
def graph_to_smiles(molecule_list: List[Tuple], atom_decoder: list) -> List[Optional[str]]:
|
50 |
+
|
51 |
+
smiles_list = []
|
52 |
+
for index, graph in enumerate(molecule_list):
|
53 |
+
try:
|
54 |
+
atom_types, edge_types = graph
|
55 |
+
mol_init = build_molecule_with_partial_charges(atom_types, edge_types, atom_decoder)
|
56 |
+
|
57 |
+
# Try to correct the molecule with connection=True, then False if needed
|
58 |
+
for connection in (True, False):
|
59 |
+
mol_conn, _ = correct_mol(mol_init, connection=connection)
|
60 |
+
if mol_conn is not None:
|
61 |
+
break
|
62 |
+
else:
|
63 |
+
logger.warning(f"Failed to correct molecule {index}")
|
64 |
+
mol_conn = mol_init # Fallback to initial molecule
|
65 |
+
|
66 |
+
# Convert to SMILES
|
67 |
+
smiles = mol2smiles(mol_conn)
|
68 |
+
if not smiles:
|
69 |
+
logger.warning(f"Failed to convert molecule {index} to SMILES, falling back to RDKit MolToSmiles")
|
70 |
+
smiles = Chem.MolToSmiles(mol_conn)
|
71 |
+
|
72 |
+
if smiles:
|
73 |
+
mol = get_mol(smiles)
|
74 |
+
if mol is not None:
|
75 |
+
# Get the largest fragment
|
76 |
+
mol_frags = Chem.rdmolops.GetMolFrags(mol, asMols=True, sanitizeFrags=False)
|
77 |
+
largest_mol = max(mol_frags, key=lambda m: m.GetNumAtoms())
|
78 |
+
|
79 |
+
largest_smiles = mol2smiles(largest_mol)
|
80 |
+
if largest_smiles and len(largest_smiles) > 1:
|
81 |
+
if check_polymer(largest_smiles):
|
82 |
+
smiles_list.append(largest_smiles)
|
83 |
+
else:
|
84 |
+
smiles_list.append(None)
|
85 |
+
elif check_polymer(smiles):
|
86 |
+
smiles_list.append(smiles)
|
87 |
+
else:
|
88 |
+
smiles_list.append(None)
|
89 |
+
else:
|
90 |
+
logger.warning(f"Failed to convert SMILES back to molecule for index {index}")
|
91 |
+
smiles_list.append(None)
|
92 |
+
else:
|
93 |
+
logger.warning(f"Failed to generate SMILES for molecule {index}, appending None")
|
94 |
+
smiles_list.append(None)
|
95 |
+
|
96 |
+
except Exception as e:
|
97 |
+
logger.error(f"Error processing molecule {index}: {str(e)}")
|
98 |
+
try:
|
99 |
+
# Fallback to RDKit's MolToSmiles if everything else fails
|
100 |
+
fallback_smiles = Chem.MolToSmiles(mol_init)
|
101 |
+
if fallback_smiles:
|
102 |
+
smiles_list.append(fallback_smiles)
|
103 |
+
logger.warning(f"Used RDKit MolToSmiles fallback for molecule {index}")
|
104 |
+
else:
|
105 |
+
smiles_list.append(None)
|
106 |
+
logger.warning(f"RDKit MolToSmiles fallback failed for molecule {index}, appending None")
|
107 |
+
except Exception as e2:
|
108 |
+
logger.error(f"All attempts failed for molecule {index}: {str(e2)}")
|
109 |
+
smiles_list.append(None)
|
110 |
+
|
111 |
+
return smiles_list
|
112 |
+
|
113 |
+
def build_molecule_with_partial_charges(
|
114 |
+
atom_types, edge_types, atom_decoder, verbose=False
|
115 |
+
):
|
116 |
+
if verbose:
|
117 |
+
print("\nbuilding new molecule")
|
118 |
+
|
119 |
+
mol = Chem.RWMol()
|
120 |
+
for atom in atom_types:
|
121 |
+
a = Chem.Atom(atom_decoder[atom.item()])
|
122 |
+
mol.AddAtom(a)
|
123 |
+
if verbose:
|
124 |
+
print("Atom added: ", atom.item(), atom_decoder[atom.item()])
|
125 |
+
|
126 |
+
edge_types = torch.triu(edge_types)
|
127 |
+
all_bonds = torch.nonzero(edge_types)
|
128 |
+
|
129 |
+
for i, bond in enumerate(all_bonds):
|
130 |
+
if bond[0].item() != bond[1].item():
|
131 |
+
mol.AddBond(
|
132 |
+
bond[0].item(),
|
133 |
+
bond[1].item(),
|
134 |
+
bond_dict[edge_types[bond[0], bond[1]].item()],
|
135 |
+
)
|
136 |
+
if verbose:
|
137 |
+
print(
|
138 |
+
"bond added:",
|
139 |
+
bond[0].item(),
|
140 |
+
bond[1].item(),
|
141 |
+
edge_types[bond[0], bond[1]].item(),
|
142 |
+
bond_dict[edge_types[bond[0], bond[1]].item()],
|
143 |
+
)
|
144 |
+
# add formal charge to atom: e.g. [O+], [N+], [S+]
|
145 |
+
# not support [O-], [N-], [S-], [NH+] etc.
|
146 |
+
flag, atomid_valence = check_valency(mol)
|
147 |
+
if verbose:
|
148 |
+
print("flag, valence", flag, atomid_valence)
|
149 |
+
if flag:
|
150 |
+
continue
|
151 |
+
else:
|
152 |
+
if len(atomid_valence) == 2:
|
153 |
+
idx = atomid_valence[0]
|
154 |
+
v = atomid_valence[1]
|
155 |
+
an = mol.GetAtomWithIdx(idx).GetAtomicNum()
|
156 |
+
if verbose:
|
157 |
+
print("atomic num of atom with a large valence", an)
|
158 |
+
if an in (7, 8, 16) and (v - ATOM_VALENCY[an]) == 1:
|
159 |
+
mol.GetAtomWithIdx(idx).SetFormalCharge(1)
|
160 |
+
# print("Formal charge added")
|
161 |
+
else:
|
162 |
+
continue
|
163 |
+
return mol
|
164 |
+
|
165 |
+
|
166 |
+
def correct_mol(mol, connection=False):
|
167 |
+
#####
|
168 |
+
no_correct = False
|
169 |
+
flag, _ = check_valency(mol)
|
170 |
+
if flag:
|
171 |
+
no_correct = True
|
172 |
+
|
173 |
+
while True:
|
174 |
+
if connection:
|
175 |
+
mol_conn = connect_fragments(mol)
|
176 |
+
mol = mol_conn
|
177 |
+
if mol is None:
|
178 |
+
return None, no_correct
|
179 |
+
flag, atomid_valence = check_valency(mol)
|
180 |
+
if flag:
|
181 |
+
break
|
182 |
+
else:
|
183 |
+
try:
|
184 |
+
assert len(atomid_valence) == 2
|
185 |
+
idx = atomid_valence[0]
|
186 |
+
v = atomid_valence[1]
|
187 |
+
queue = []
|
188 |
+
check_idx = 0
|
189 |
+
for b in mol.GetAtomWithIdx(idx).GetBonds():
|
190 |
+
type = int(b.GetBondType())
|
191 |
+
queue.append(
|
192 |
+
(b.GetIdx(), type, b.GetBeginAtomIdx(), b.GetEndAtomIdx())
|
193 |
+
)
|
194 |
+
if type == 12:
|
195 |
+
check_idx += 1
|
196 |
+
queue.sort(key=lambda tup: tup[1], reverse=True)
|
197 |
+
|
198 |
+
if queue[-1][1] == 12:
|
199 |
+
return None, no_correct
|
200 |
+
elif len(queue) > 0:
|
201 |
+
start = queue[check_idx][2]
|
202 |
+
end = queue[check_idx][3]
|
203 |
+
t = queue[check_idx][1] - 1
|
204 |
+
mol.RemoveBond(start, end)
|
205 |
+
if t >= 1:
|
206 |
+
mol.AddBond(start, end, bond_dict[t])
|
207 |
+
except Exception as e:
|
208 |
+
# print(f"An error occurred in correction: {e}")
|
209 |
+
return None, no_correct
|
210 |
+
return mol, no_correct
|
211 |
+
|
212 |
+
def check_valid(smiles):
|
213 |
+
mol = get_mol(smiles)
|
214 |
+
if mol is None:
|
215 |
+
return False
|
216 |
+
smiles = mol2smiles(mol)
|
217 |
+
if smiles is None:
|
218 |
+
return False
|
219 |
+
return True
|
220 |
+
|
221 |
+
def get_mol(smiles_or_mol):
|
222 |
+
"""
|
223 |
+
Loads SMILES/molecule into RDKit's object
|
224 |
+
"""
|
225 |
+
if isinstance(smiles_or_mol, str):
|
226 |
+
if len(smiles_or_mol) == 0:
|
227 |
+
return None
|
228 |
+
mol = Chem.MolFromSmiles(smiles_or_mol)
|
229 |
+
if mol is None:
|
230 |
+
return None
|
231 |
+
try:
|
232 |
+
Chem.SanitizeMol(mol)
|
233 |
+
except ValueError:
|
234 |
+
return None
|
235 |
+
return mol
|
236 |
+
return smiles_or_mol
|
237 |
+
|
238 |
+
|
239 |
+
def mol2smiles(mol):
|
240 |
+
if mol is None:
|
241 |
+
return None
|
242 |
+
try:
|
243 |
+
Chem.SanitizeMol(mol)
|
244 |
+
except ValueError:
|
245 |
+
return None
|
246 |
+
return Chem.MolToSmiles(mol)
|
247 |
+
|
248 |
+
|
249 |
+
def check_valency(mol):
|
250 |
+
try:
|
251 |
+
# First attempt to sanitize with specific properties
|
252 |
+
Chem.SanitizeMol(mol, sanitizeOps=Chem.SanitizeFlags.SANITIZE_PROPERTIES)
|
253 |
+
return True, None
|
254 |
+
except ValueError as e:
|
255 |
+
e = str(e)
|
256 |
+
p = e.find("#")
|
257 |
+
e_sub = e[p:]
|
258 |
+
atomid_valence = list(map(int, re.findall(r"\d+", e_sub)))
|
259 |
+
return False, atomid_valence
|
260 |
+
except Exception as e:
|
261 |
+
# print(f"An unexpected error occurred: {e}")
|
262 |
+
return False, []
|
263 |
+
|
264 |
+
|
265 |
+
##### connect fragements
|
266 |
+
def select_atom_with_available_valency(frag):
|
267 |
+
atoms = list(frag.GetAtoms())
|
268 |
+
random.shuffle(atoms)
|
269 |
+
for atom in atoms:
|
270 |
+
if atom.GetAtomicNum() > 1 and atom.GetImplicitValence() > 0:
|
271 |
+
return atom
|
272 |
+
return None
|
273 |
+
|
274 |
+
|
275 |
+
def select_atoms_with_available_valency(frag):
|
276 |
+
return [
|
277 |
+
atom
|
278 |
+
for atom in frag.GetAtoms()
|
279 |
+
if atom.GetAtomicNum() > 1 and atom.GetImplicitValence() > 0
|
280 |
+
]
|
281 |
+
|
282 |
+
|
283 |
+
def try_to_connect_fragments(combined_mol, frag, atom1, atom2):
|
284 |
+
# Make copies of the molecules to try the connection
|
285 |
+
trial_combined_mol = Chem.RWMol(combined_mol)
|
286 |
+
trial_frag = Chem.RWMol(frag)
|
287 |
+
|
288 |
+
# Add the new fragment to the combined molecule with new indices
|
289 |
+
new_indices = {
|
290 |
+
atom.GetIdx(): trial_combined_mol.AddAtom(atom)
|
291 |
+
for atom in trial_frag.GetAtoms()
|
292 |
+
}
|
293 |
+
|
294 |
+
# Add the bond between the suitable atoms from each fragment
|
295 |
+
trial_combined_mol.AddBond(
|
296 |
+
atom1.GetIdx(), new_indices[atom2.GetIdx()], Chem.BondType.SINGLE
|
297 |
+
)
|
298 |
+
|
299 |
+
# Adjust the hydrogen count of the connected atoms
|
300 |
+
for atom_idx in [atom1.GetIdx(), new_indices[atom2.GetIdx()]]:
|
301 |
+
atom = trial_combined_mol.GetAtomWithIdx(atom_idx)
|
302 |
+
num_h = atom.GetTotalNumHs()
|
303 |
+
atom.SetNumExplicitHs(max(0, num_h - 1))
|
304 |
+
|
305 |
+
# Add bonds for the new fragment
|
306 |
+
for bond in trial_frag.GetBonds():
|
307 |
+
trial_combined_mol.AddBond(
|
308 |
+
new_indices[bond.GetBeginAtomIdx()],
|
309 |
+
new_indices[bond.GetEndAtomIdx()],
|
310 |
+
bond.GetBondType(),
|
311 |
+
)
|
312 |
+
|
313 |
+
# Convert to a Mol object and try to sanitize it
|
314 |
+
new_mol = Chem.Mol(trial_combined_mol)
|
315 |
+
try:
|
316 |
+
Chem.SanitizeMol(new_mol)
|
317 |
+
return new_mol # Return the new valid molecule
|
318 |
+
except Chem.MolSanitizeException:
|
319 |
+
return None # If the molecule is not valid, return None
|
320 |
+
|
321 |
+
|
322 |
+
def connect_fragments(mol):
|
323 |
+
# Get the separate fragments
|
324 |
+
frags = Chem.GetMolFrags(mol, asMols=True, sanitizeFrags=False)
|
325 |
+
if len(frags) < 2:
|
326 |
+
return mol
|
327 |
+
|
328 |
+
combined_mol = Chem.RWMol(frags[0])
|
329 |
+
|
330 |
+
for frag in frags[1:]:
|
331 |
+
# Select all atoms with available valency from both molecules
|
332 |
+
atoms1 = select_atoms_with_available_valency(combined_mol)
|
333 |
+
atoms2 = select_atoms_with_available_valency(frag)
|
334 |
+
|
335 |
+
# Try to connect using all combinations of available valency atoms
|
336 |
+
for atom1 in atoms1:
|
337 |
+
for atom2 in atoms2:
|
338 |
+
new_mol = try_to_connect_fragments(combined_mol, frag, atom1, atom2)
|
339 |
+
if new_mol is not None:
|
340 |
+
# If a valid connection is made, update the combined molecule and break
|
341 |
+
combined_mol = new_mol
|
342 |
+
break
|
343 |
+
else:
|
344 |
+
# Continue if the inner loop didn't break (no valid connection found for atom1)
|
345 |
+
continue
|
346 |
+
# Break if the inner loop did break (valid connection found)
|
347 |
+
break
|
348 |
+
else:
|
349 |
+
# If no valid connections could be made with any of the atoms, return None
|
350 |
+
return None
|
351 |
+
|
352 |
+
return combined_mol
|
353 |
+
|
354 |
+
|
355 |
+
#### connect fragements
|