-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathfragmentation.py
159 lines (135 loc) · 6.02 KB
/
fragmentation.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
import os
import argparse
import yaml
from easydict import EasyDict
from glob import glob
import torch
from utils.frag_utils import *
from utils.data_utils import *
from utils.misc import *
import pickle
import rdkit.Chem as Chem
from torch_geometric.utils import contains_isolated_nodes
def cluster_matrix(assignment, fg_size):
assign_mat = torch.zeros(size=[fg_size, len(assignment)])
for i in range(len(assignment)):
assign_mat[assignment[i], i] = 1
assign_mat = torch.t(assign_mat)
assign_mat = assign_mat / (torch.sum(assign_mat, 1).unsqueeze(-1))
assign_mat = assign_mat.to_sparse()
return assign_mat
def vocab_gen_ps(smis, rdmols, vocab_len, vocab_path, cpus, kekulize):
# loop
mols = [MolInSubgraph(mol, kekulize) for mol in rdmols]
selected_smis, details = list(MAX_VALENCE.keys()), {} # details: <smi: [atom cnt, frequency]
# calculate single atom frequency
for atom in selected_smis:
details[atom] = [1, 0] # frequency of single atom is not calculated
for smi in smis:
cnts = cnt_atom(smi, return_dict=True)
for atom in details:
if atom in cnts:
details[atom][1] += cnts[atom]
# bpe process
add_len = vocab_len - len(selected_smis)
# pbar = tqdm(total=add_len)
pool = mp.Pool(cpus)
while len(selected_smis) < vocab_len:
res_list = pool.map(freq_cnt, mols) # each element is (freq, mol) (because mol will not be synced...)
freqs, mols = {}, []
for freq, mol in res_list:
mols.append(mol)
for key in freq:
freqs.setdefault(key, 0)
freqs[key] += freq[key]
# find the subgraph to merge
max_cnt, merge_smi = 0, ''
for smi in freqs:
cnt = freqs[smi]
if cnt > max_cnt:
max_cnt = cnt
merge_smi = smi
# merge
for mol in mols:
mol.merge(merge_smi)
if merge_smi in details: # corner case: re-extracted from another path
continue
selected_smis.append(merge_smi)
details[merge_smi] = [cnt_atom(merge_smi), max_cnt]
# pbar.update(1)
# pbar.close()
print_log('sorting vocab by atom num')
selected_smis.sort(key=lambda x: details[x][0], reverse=True)
pool.close()
with open(vocab_path, 'w') as fout:
fout.write(json.dumps({'kekulize': kekulize}) + '\n')
fout.writelines(list(map(lambda smi: f'{smi}\t{details[smi][0]}\t{details[smi][1]}\n', selected_smis)))
return selected_smis, details
def fragment_ps(tokenizer, mols, vocab_len, indices):
train_end_idx, val_end_idx = indices[0], indices[1]
for idx, mol in enumerate(mols):
n_atoms = mol.GetNumAtoms()
if idx < train_end_idx:
graph = train_set.data[idx]
elif train_end_idx <= idx < val_end_idx:
graph = val_set.data[idx-train_end_idx]
elif val_end_idx <= idx:
graph = test_set.data[idx-val_end_idx]
frag_list = tokenizer.tokenize(mol)
cg, cg_smiles = [], []
for frag in frag_list:
cg_smiles.append(frag[0])
cg.append(frag[1])
fg2cg = cluster_matrix(cg, n_atoms)
cg_pos = torch.matmul(fg2cg, graph.pos)
graph.fg2cg = cg
graph.cg_pos = cg_pos
graph.cg_smile = cg_smiles
graph.num_frags = len(cg)
train_data_path = os.path.join(os.path.dirname(vocab_path), 'train_data_40k_ps'+str(vocab_len)+'.pkl')
val_data_path = os.path.join(os.path.dirname(vocab_path), 'val_data_5k_ps'+str(vocab_len)+'.pkl')
test_data_path = os.path.join(os.path.dirname(vocab_path), 'test_data_1k_ps'+str(vocab_len)+'.pkl')
with open(train_data_path, "wb") as fout:
pickle.dump(train_set.data, fout)
with open(val_data_path, 'wb') as fout:
pickle.dump(val_set.data, fout)
with open(test_data_path, 'wb') as fout:
pickle.dump(test_set.data, fout)
return train_set, val_set, test_set
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('config', type=str)
parser.add_argument('--device', type=str, default='cuda')
parser.add_argument('--vocab_len', type=int, default=500, help='size of vocab')
parser.add_argument('--cpus', type=int, default=16)
parser.add_argument('--kekulize', type=bool, default=False)
args = parser.parse_args()
resume = os.path.isdir(args.config)
if resume:
config_path = glob(os.path.join(args.config, '*.yml'))[0]
resume_from = args.config
else:
config_path = args.config
with open(config_path, 'r') as f:
config = EasyDict(yaml.safe_load(f))
config_name = os.path.basename(config_path)[:os.path.basename(config_path).rfind('.')]
seed_all(config.train.seed)
# Loading dataset
transforms = CountNodesPerGraph()
test_transforms = Compose([CountNodesPerGraph()])
train_set = ConformationDataset(config.dataset.train, transform=transforms)
val_set = ConformationDataset(config.dataset.val, transform=transforms)
test_set = PackedConformationDataset(config.dataset.test, transform=test_transforms)
train_end_idx, val_end_idx = train_set.__len__(), train_set.__len__()+val_set.__len__()
data_indices = [train_end_idx, val_end_idx]
smis = [Chem.MolToSmiles(train_set.data[i].rdmol) for i in range(len(train_set))] + \
[Chem.MolToSmiles(val_set.data[i].rdmol) for i in range(len(val_set))] + \
[Chem.MolToSmiles(test_set.data[i].rdmol) for i in range(len(test_set.data))]
mols = [train_set.data[i].rdmol for i in range(len(train_set))] + \
[val_set.data[i].rdmol for i in range(len(val_set))] + \
[test_set.data[i].rdmol for i in range(len(test_set.data))]
# Vocabulary generation
vocab_path = os.path.join(os.path.dirname(config.dataset.train), 'vocab_'+args.frag+str(args.vocab_len)+'.pkl')
vocab_gen_ps(smis=smis, rdmols=mols, vocab_path=vocab_path, vocab_len=args.vocab_len, cpus=args.cpus, kekulize=args.kekulize)
tokenizer = Tokenizer(vocab_path)
fragment_ps(tokenizer=tokenizer, mols=mols, vocab_len=args.vocab_len, indices=data_indices)